Align the SQLite run store test with honest step failures

Why: the U3B semantics change removed the failed-with-patches promotion; TestEngineRunsOnSQLiteStore still encoded the old behavior and failed under the fixed engine. Evidence: repair receipt success; sqlite store tests pass (verify-u3br2.json). Impact: test-only change. Refs: feature/agent-stack-audit-remediation-b1 U3B repair
Pin nested agent-intent and agent-native to the batch-1 fixes
2026-10-21 18:05:51 +00:00 · 2026-10-21 17:45:01 +00:00 · 2026-10-21 17:20:40 +00:00 · 2026-10-21 16:44:24 +00:00 · 2026-10-21 15:51:53 +00:00 · 2026-10-20 08:57:38 +00:00
42 changed files with 3274 additions and 108 deletions
@@ -0,0 +1,150 @@
+name: governance
+on:
+  pull_request:
+  push:
+    branches: [main]
+
+jobs:
+  governance:
+    runs-on: ubuntu-latest
+    env:
+      GOPROXY: https://goproxy.cn,direct
+    steps:
+      - name: Checkout
+        env:
+          TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          SUBMODULE_TOKEN: ${{ secrets.SUBMODULE_TOKEN }}
+        run: |
+          set -euo pipefail
+          # 只物化两个治理 gitlink（Guidelines 与 deps/development-manager），非递归；
+          # 其它 gitlink 不物化、不取回。凭据只以 URL 限定的 extraheader 存在于本进程树内。
+          python3 - <<'PY_CHECKOUT'
+          import os
+          from pathlib import Path
+          import re
+          import subprocess
+          import sys
+          from urllib.parse import urlsplit
+
+          # 服务端基址取运行器注入的 GITHUB_SERVER_URL（平台控制、候选 PR 不可影响）。
+          SERVER = os.environ.get("GITHUB_SERVER_URL", "").rstrip("/")
+          parsed_server = urlsplit(SERVER)
+          if parsed_server.scheme not in {"http", "https"} or not parsed_server.netloc:
+              sys.exit("Checkout server must be the platform-injected http(s) URL")
+          SSH = "ssh://git@git-ssh.u1680832.nyat.app:31411/weichen/"
+          # 审查过的白名单，独立于候选 .gitmodules：仅列出两个治理 gitlink。
+          REPOSITORIES = {
+              "deps/development-guidelines": "Development-Guidelines",
+              "deps/development-manager": "Development-Manager",
+          }
+          TARGET = "weichen/Agent-Squad"
+          root_token = os.environ.pop("TOKEN", "")
+          component_token = os.environ.pop("SUBMODULE_TOKEN", "")
+          if not root_token or not component_token:
+              sys.exit("Checkout requires TOKEN and SUBMODULE_TOKEN")
+          if os.environ.get("GITHUB_REPOSITORY") != TARGET:
+              sys.exit("Checkout repository is outside the reviewed allowlist")
+          sha = os.environ.get("GITHUB_SHA", "")
+          if not re.fullmatch(r"[0-9a-fA-F]{40}", sha):
+              sys.exit("Checkout requires a fixed commit SHA")
+
+          # 丢弃继承的 Git 追踪/配置/SSH 钩子；凭据只存在于单个 Git 进程树的精确 URL 头中。
+          clean_env = {k: v for k, v in os.environ.items() if not k.startswith("GIT_")}
+          clean_env.update(GIT_CONFIG_NOSYSTEM="1", GIT_CONFIG_SYSTEM="/dev/null",
+                           GIT_CONFIG_GLOBAL="/dev/null", GIT_TERMINAL_PROMPT="0")
+          config = {
+              "core.hooksPath": "/dev/null", "core.fsmonitor": "false",
+              "init.templateDir": "", "credential.helper": "",
+              "fetch.recurseSubmodules": "false", "submodule.recurse": "false",
+              "http.followRedirects": "false", "protocol.allow": "never",
+              "protocol." + urlsplit(SERVER).scheme + ".allow": "always",
+          }
+
+
+          def git(repo, *args, extra=None):
+              values = config | (extra or {})
+              env = clean_env | {"GIT_CONFIG_COUNT": str(len(values))}
+              for index, (key, value) in enumerate(values.items()):
+                  env[f"GIT_CONFIG_KEY_{index}"] = key
+                  env[f"GIT_CONFIG_VALUE_{index}"] = value
+              return subprocess.run(["git", "-C", str(repo), *args], env=env,
+                                    check=True, stdout=subprocess.PIPE).stdout
+
+
+          def materialize_governance(root):
+              # -r 列出任意深度的 gitlink，但不进入 submodule 内部；其它 gitlink 只记录、从不取回。
+              links = {}
+              for entry in git(root, "ls-tree", "-rz", "HEAD").split(b"\0"):
+                  if not entry:
+                      continue
+                  metadata, path = entry.split(b"\t", 1)
+                  mode, kind, commit = metadata.decode().split()
+                  if mode == "160000":
+                      links[path.decode()] = commit
+              for name in REPOSITORIES:
+                  if name not in links:
+                      raise ValueError("Governance gitlink is missing from the fixed commit")
+              raw = git(root, "config", "--no-includes", "--blob", "HEAD:.gitmodules", "--null", "--list")
+              entries = {}
+              for item in raw.split(b"\0"):
+                  if not item:
+                      continue
+                  key, value = item.decode().split("\n", 1)
+                  match = re.fullmatch(r"submodule\.(.+)\.(path|url)", key)
+                  if not match:
+                      continue
+                  name, field = match.groups()
+                  if name not in REPOSITORIES:
+                      continue
+                  if field in entries.setdefault(name, {}):
+                      raise ValueError("Duplicate candidate .gitmodules setting")
+                  entries[name][field] = value
+              # 先整体校验两个治理条目，再进行任何带凭据的取回。
+              for name, repository in REPOSITORIES.items():
+                  fields = entries.get(name, {})
+                  if fields.get("path") != name:
+                      raise ValueError("Governance submodule path is outside the reviewed allowlist")
+                  trusted_url = f"{SERVER}/weichen/{repository}.git"
+                  if fields.get("url") not in {SSH + repository + ".git", trusted_url}:
+                      raise ValueError("Governance submodule URL is outside the reviewed allowlist")
+              for name, repository in REPOSITORIES.items():
+                  url = f"{SERVER}/weichen/{repository}.git"
+                  # 仅持久化审查过的公开 URL，使 Git 在进程级配置消失后仍识别该 submodule。
+                  git(root, "config", f"submodule.{name}.url", url)
+                  git(root, "submodule", "update", "--init", "--checkout", "--", name,
+                      extra={f"submodule.{name}.url": url,
+                             f"submodule.{name}.update": "checkout",
+                             f"http.{url}.extraheader": "AUTHORIZATION: token " + component_token})
+                  child = root / name
+                  if git(child, "rev-parse", "HEAD").decode().strip() != links[name]:
+                      raise ValueError("Materialized governance gitlink does not match the fixed SHA")
+
+
+          try:
+              root = Path.cwd()
+              # 必须是干净的临时工作区；不复用既有 Git 状态。
+              if (root / ".git").exists():
+                  raise ValueError("Checkout requires a fresh workspace")
+              git(root, "init", "-q")
+              url = f"{SERVER}/{TARGET}.git"
+              git(root, "remote", "add", "origin", url)
+              git(root, "fetch", "--depth=1", "origin", sha,
+                  extra={f"http.{url}.extraheader": "AUTHORIZATION: token " + root_token})
+              git(root, "checkout", "-q", "--detach", "FETCH_HEAD")
+              materialize_governance(root)
+          except (ValueError, OSError, subprocess.CalledProcessError) as error:
+              sys.exit(f"Checkout refused: {error}")
+          print("Checkout complete; credentials were not persisted in Git configuration")
+          PY_CHECKOUT
+          unset TOKEN SUBMODULE_TOKEN
+      - name: Install Go 1.27.1
+        run: |
+          set -euo pipefail
+          curl -fsSL -o "$RUNNER_TEMP/go.tgz" https://mirrors.aliyun.com/golang/go1.27.1.linux-amd64.tar.gz
+          tar -C /usr/local -xzf "$RUNNER_TEMP/go.tgz"
+          ln -sfn /usr/local/go/bin/go /usr/local/bin/go
+          ln -sfn /usr/local/go/bin/gofmt /usr/local/bin/gofmt
+      - name: Validate agent execution configuration
+        run: scripts/agent-exec validate --repo .
+      - name: Governance check
+        run: scripts/dm governance check --repo .
@@ -13,3 +13,6 @@
 [submodule "deps/agent-intent"]
 	path = deps/agent-intent
 	url = ssh://git@git-ssh.u1680832.nyat.app:31411/weichen/Agent-Intent.git
+[submodule "deps/development-manager"]
+	path = deps/development-manager
+	url = ssh://git@git-ssh.u1680832.nyat.app:31411/weichen/Development-Manager.git
@@ -0,0 +1,58 @@
+# Agent-Squad 采纳 development-governance-v2 治理套件 任务卡
+
+任务主记录：[Agent-Squad PR #54](https://git-home.u1680832.nyat.app:30319/weichen/Agent-Squad/pulls/54)（本文件随 PR 描述维护）。
+状态：已完成（闭合随本 PR 合入；合入以平台记录为准）；owner：weichen。
+
+## 基线与授权
+
+- 权威仓：weichen/Agent-Squad；root_target：main；immediate_target：main
+- branch_kind：initiative；initiative_id：governance-adoption-Agent-Squad
+- R0（initial_main_sha）：490b0b2dfa1567cb16ff7a7230fd78e52bc2938e；工作区：/home/weichen/projects/Agent-Squad-wt-feature-governance-adoption（worktree）；分支：feature/governance-adoption
+- 父任务信息（仅子任务填写）：无
+- accepting_requirements：false（单仓采纳 initiative，范围固定为下列十步，不接收新需求）
+- 需求分流：verdict=NEW；selected_initiative=无；候选=无；理由=development-governance-v2 Phase 6 每个正式仓库独立 initiative；triage 收据=无（本仓采纳前尚无 triage 入口）
+- 授权与交付终点：owner 已授权 development-governance-v2 全部实施（Agent-System-Dev 主记录 docs/superpowers/plans/2026-10-16-development-governance-v2-master.md）；终点为本 PR 合入 main 且采纳状态 active。分支保护与 tag 由用户执行。
+- 实施执行：agent-per-unit（配置：AGENT-EXECUTION.md；外部 Agent 一单元一新进程/会话/工作区）
+
+## 目标与非目标
+
+目标（Phase 6 十步）：固定 Guidelines SHA（deps/development-guidelines @ 2688fd2be6a52c0bf110e7e6366b6f291ab5e4eb）；增加 AGENT-EXECUTION.md；增加共享执行器薄入口（deps/development-manager @ bd3a7de57a3a9e1a4495b94c5fa5c08d71e7be8c + scripts/dm、scripts/agent-exec、scripts/task）；四条路由（implementation/repair/test/review）；任务卡模板与分支元数据；CI 中启用 schema 校验（.gitea/workflows/governance.yml：agent-exec validate + governance check）；一次正例运行；负例（缺配置、坏收据、越界路径、未知副作用）；独立 Review + CI；采纳状态 active。
+
+非目标：不改业务代码与既有测试；不改分支保护（由用户执行）；不发布 tag。
+
+范围说明（Review 阻塞项 2）：`scripts/submodule-consistency.sh` 要求 deps/agent-native、deps/agent-intent 内嵌的 Guidelines/错误码/日志 gitlink 与本仓一致；本 PR 把 Guidelines 固定到 2688fd2 后，按最小修复把这两个 gitlink 升到各自已合入治理采纳的 main（Agent-Native 14d4869、Agent-Intent 651a2e1；两者相对原固定版本只增治理文件、不改 Go 代码），其余 gitlink 不动。核对见证据 `scripts/submodule-consistency.sh`。
+
+引导说明：AGENT-EXECUTION.md 由 host 以固定版本治理套件（dm-server governance init @ bd3a7de57a3a9e1a4495b94c5fa5c08d71e7be8c）生成并作为引导配置提交（执行器运行前必须存在有效配置；沿用 DM 采纳先例）；其余文件由实施单元生成。
+
+## Agent 实施收据
+
+| 单元 ID | 预期范围 | 角色 | 候选顺序 | attempt | CLI/模型 | 终态 | transport | fallback | 源码外收据路径 |
+|---|---|---|---|---|---|---|---|---|---|
+| governance-adoption-Agent-Squad-r1 | 13 个允许路径（治理文件 + 两个 gitlink） |  | implementation | cursor→antigravity→pi | run 71696608（dm-server pinned，attempts=1） | cursor / not-observable | failed-unknown | 无（apply 快进） | 否 | /tmp/gov-receipt-Agent-Squad-r1.json |
+| 对账（r1） | 规范 worktree 仍为 2a2a1fd 且干净；无 transport ref | — | — | — | — | — | 已对账；全新运行 r2（单元包追加禁止固定共享 /tmp 路径约束） | — | — | — |
+| governance-adoption-Agent-Squad-r2 | 13 个允许路径（治理文件 + 两个 gitlink） | b83905a8 | implementation | cursor→antigravity→pi | run 145e30ce（dm-server 42e675f，attempts=1） | cursor / not-observable | success | 468878f62ede（apply 快进） | 否 | /tmp/gov-receipt-Agent-Squad-r2.json |
+| （无效启动 repair-r0） | — | repair | — | 无 attempt | — | invalid-config（单元包尚未写出即启动，执行器读取失败；无 attempt、无副作用，worktree 仍为 e29284e 且干净） | 无 | 否 | /tmp/gov-receipt-Agent-Squad-repair-r0-invalid-unit-path.json |
+| governance-adoption-Agent-Squad-repair-r1 | policy/guidelines-adoption.md、PROJECT.md、deps/agent-native、deps/agent-intent | repair | cursor→antigravity→pi | run fadb9c92（attempts=1） | cursor / not-observable | success | d0f9081f47eb（apply 快进） | 否 | /tmp/gov-receipt-Agent-Squad-repair-r1.json |
+
+## 同步记录
+
+| 节点 | sync_base_sha | tested_head_sha | checked_at | 结果 |
+|---|---|---|---|---|
+| 启动 | 490b0b2dfa1567cb16ff7a7230fd78e52bc2938e | 490b0b2dfa1567cb16ff7a7230fd78e52bc2938e | 2026-10-17T07:17:41Z | 一致 |
+
+## 结果与验证（完成后填写）
+
+- 结果：治理套件已采纳——Guidelines 与 Development-Manager gitlink 固定、AGENT-EXECUTION.md 四条路由、共享执行器薄入口、pre-commit admit、governance CI；采纳状态 active（随合入生效）。
+- 验证：见下表（命令输出由 /tmp/gov-neg/ev.sh 写入 .task/evidence/governance-adoption/，SELF 除非另注）。
+- 影响与剩余项：不改业务代码；本地启用钩子需由人执行 `git config core.hooksPath scripts/hooks`；将 governance 检查设为分支保护必需项属用户双钥匙事项（命令见 Agent-System-Dev 治理主记录）。
+
+| 时间(UTC) | 命令 | 被测 HEAD | exit | 来源 | 摘要/位置 |
+|---|---|---|---|---|---|
+| 20261017T080150Z | `scripts/dm governance check --repo .` | 468878f62ede | 0 | SELF | ok fixtures；[输出](../evidence/governance-adoption/00-20261017T080150Z.log) sha256:b07d207012a2fd6d |
+| 20261017T080150Z | `scripts/agent-exec validate --repo .` | 468878f62ede | 0 | SELF | validate: ok；[输出](../evidence/governance-adoption/01-20261017T080150Z.log) sha256:b3a91246221bb5f8 |
+| 20261017T080150Z | `python3 /tmp/gov-neg/adopt_cases.py .` | 468878f62ede | 0 | SELF | [verdict] ALL PASS cases=5；[输出](../evidence/governance-adoption/02-20261017T080150Z.log) sha256:109906d5e57fe039 |
+| 20261017T094306Z | `scripts/dm governance check --repo .` | d0f9081f47eb | 0 | SELF | ok fixtures；[输出](../evidence/governance-adoption/03-20261017T094306Z.log) sha256:b904686133da5f80 |
+| 20261017T094306Z | `scripts/agent-exec validate --repo .` | d0f9081f47eb | 0 | SELF | validate: ok；[输出](../evidence/governance-adoption/04-20261017T094306Z.log) sha256:19dc223efda318d6 |
+| 20261017T094308Z | `scripts/submodule-consistency.sh` | 05b509827274 | 0 | SELF | ok deps/agent-intent 651a2e1e03c4326888ef9f742ce592140ae30ab8；[输出](../evidence/governance-adoption/05-20261017T094308Z.log) sha256:bc528fe029ed41d2 |
+| — | 候选 HEAD CI | b3e7da4a11b9 | 0 | GATE | success ci / docker-checkout-hostgateway (pull_request)=success（/weichen/Agent-Squad/actions/runs/3672/jobs/5787）；governance / governance (pull_request)=success（/weichen/Agent-Squad/actions/runs/3673/jobs/5788）；ci / vet-and-test (pull_request)=success（/weichen/Agent-Squad/actions/runs/3672/jobs/5786） |
+| — | 独立 Review | b3e7da4a11b9 | 0 | REPLAY | cursor-agent-stable --mode ask 全新只读会话（fresh_session=true，session id not-observable）闭合重审 VERDICT: APPROVE（对象 b3e7da4） |
@@ -0,0 +1,19 @@
+command: scripts/dm governance check --repo . 
+cwd: /home/weichen/projects/Agent-Squad-wt-feature-governance-adoption
+head: 468878f62eded8347d18a1d6c1b7c5f7e24a9c4c
+exit: 0
+utc: 20261017T080150Z
+---
+ok gitlink deps/development-guidelines
+ok gitlink deps/development-manager
+ok agent-execution
+ok file AGENTS.md
+ok file PROJECT.md
+ok file policy/guidelines-adoption.md
+ok file .task/templates/task-card.md
+ok executable scripts/agent-exec
+ok executable scripts/hooks/pre-commit
+ok executable scripts/dm
+ok executable scripts/task
+ok adoption status
+ok fixtures
@@ -0,0 +1,7 @@
+command: scripts/agent-exec validate --repo . 
+cwd: /home/weichen/projects/Agent-Squad-wt-feature-governance-adoption
+head: 468878f62eded8347d18a1d6c1b7c5f7e24a9c4c
+exit: 0
+utc: 20261017T080150Z
+---
+validate: ok
@@ -0,0 +1,13 @@
+command: python3 /tmp/gov-neg/adopt_cases.py . 
+cwd: /home/weichen/projects/Agent-Squad-wt-feature-governance-adoption
+head: 468878f62eded8347d18a1d6c1b7c5f7e24a9c4c
+exit: 0
+utc: 20261017T080150Z
+---
+[object] Agent-Squad-wt-feature-governance-adoption HEAD=468878f62eded8347d18a1d6c1b7c5f7e24a9c4c
+PASS positive-run-apply: run exit=0 terminal=success apply exit=0 head=312d80d409a1
+PASS missing-config: run exit=2 terminal=invalid-config validate exit=2
+PASS bad-receipt: apply exit=4 head_unchanged=True msg=apply: no local integrity record: changed paths do not match the transport diff
+PASS out-of-scope-path: run exit=5 terminal=scope-expansion transport=False
+PASS unknown-side-effect: run exit=4 terminal=failed-unknown transport=False
+[verdict] ALL PASS cases=5
@@ -0,0 +1,19 @@
+command: scripts/dm governance check --repo . 
+cwd: /home/weichen/projects/Agent-Squad-wt-feature-governance-adoption
+head: d0f9081f47eb59565323f731cb8085e5891dd7ba
+exit: 0
+utc: 20261017T094306Z
+---
+ok gitlink deps/development-guidelines
+ok gitlink deps/development-manager
+ok agent-execution
+ok file AGENTS.md
+ok file PROJECT.md
+ok file policy/guidelines-adoption.md
+ok file .task/templates/task-card.md
+ok executable scripts/agent-exec
+ok executable scripts/hooks/pre-commit
+ok executable scripts/dm
+ok executable scripts/task
+ok adoption status
+ok fixtures
@@ -0,0 +1,7 @@
+command: scripts/agent-exec validate --repo . 
+cwd: /home/weichen/projects/Agent-Squad-wt-feature-governance-adoption
+head: d0f9081f47eb59565323f731cb8085e5891dd7ba
+exit: 0
+utc: 20261017T094306Z
+---
+validate: ok
@@ -0,0 +1,14 @@
+command: scripts/submodule-consistency.sh 
+cwd: /home/weichen/projects/Agent-Squad-wt-feature-governance-adoption
+head: 05b5098272741987767efa2e78b8fe01aa84b6e5
+exit: 0
+utc: 20261017T094308Z
+---
+ok development-guidelines@deps/agent-native 2688fd2be6a52c0bf110e7e6366b6f291ab5e4eb
+ok error-code-system@deps/agent-native a3ee67bba269e1276c14e22c026171ecf2cf6190
+ok logging-system@deps/agent-native 49b5d8be604d05e4246dffa0d10fa1c2b7ad45d7
+ok development-guidelines@deps/agent-intent 2688fd2be6a52c0bf110e7e6366b6f291ab5e4eb
+ok error-code-system@deps/agent-intent a3ee67bba269e1276c14e22c026171ecf2cf6190
+ok logging-system@deps/agent-intent 49b5d8be604d05e4246dffa0d10fa1c2b7ad45d7
+ok deps/agent-native 14d48697a69ddba9ebc67ba7daa588f1b082d096
+ok deps/agent-intent 651a2e1e03c4326888ef9f742ce592140ae30ab8
@@ -0,0 +1,66 @@
+# <任务标题> 任务卡
+
+任务主记录：<Issue/PR 链接，或注明"本文件随 PR 描述维护">。
+状态：<进行中 / 待审核 / 已合入 / 已放弃>；owner：<填写>。
+
+## 基线与授权
+
+- 权威仓：<仓 URL 或逻辑 ID>；root_target：main；immediate_target：<main 或 feature/initiative>
+- branch_kind：<initiative | child | standalone>；initiative_id：<稳定 ID，普通任务为空>
+- R0（initial_main_sha）：<完整 SHA>；工作区：<路径>（worktree / clone）；分支：<feature|fix|hotfix/...>
+- 父任务信息（仅子任务填写）：parent_branch=<父分支>；parent_head_sha=<SHA>；parent_pr=<父 PR>
+- accepting_requirements：<true | false>（仅 initiative 可为 true；冻结后为 false）
+- 需求分流：verdict=<JOIN | NEW>；selected_initiative=<分支或无>；候选=<分支@HEAD 列表或无>；理由=<不适配/适配依据>；triage 收据=<源码外路径与摘要>
+- 授权与交付终点：<本次允许的动作与终点；未授权事项写明>
+- 实施执行：agent-per-unit（配置：AGENT-EXECUTION.md；外部 Agent 一单元一新进程/会话/工作区）
+
+## 目标与非目标
+
+<预期结果、受影响入口/模块/契约；明确不做什么。>
+
+## Agent 实施收据
+
+按根目录 AGENT-EXECUTION.md 路由；外部 Agent 一单元一新进程/会话/隔离工作区。实施、修复与测试必须通过登记的候选执行；未配置或候选安全耗尽停在 unavailable/blocked，禁止 host 隐式代写。
+
+| 单元 ID | 预期范围 | 角色 | 候选顺序 | attempt | CLI/模型 | 终态 | transport | fallback | 源码外收据路径 |
+|---|---|---|---|---|---|---|---|---|---|
+| 待定义 | — | NOT_RUN | NOT_RUN | NOT_RUN | NOT_RUN | NOT_RUN | NOT_RUN | NOT_RUN | — |
+
+## 复用材料（task-start 自动检索结果摘录）
+
+- 相关近期提交：<git log -- <paths> 命中；无则写无>。
+- 相关经验条目：<检索命中；无则写无>。
+
+## 同步记录
+
+| 节点 | sync_base_sha | tested_head_sha | checked_at | 结果 |
+|---|---|---|---|---|
+| 启动 | <R0> | <HEAD> | <日期时间> | <一致/已同步> |
+| 验收前 | <填写> | <填写> | <填写> | <填写> |
+
+字段没有发生的写"无"，不省略。本卡与 PR 描述同源：开 PR 时将"目标与非目标、结果与验证、影响与剩余项"复制进 PR 描述，不另写第二份。
+
+## 交付调度（有多个就绪任务时）
+
+- 调度记录 / entry_id / 当前 attempt_id：<平台或调度服务链接及标识；未入队写无>。
+- 入队源 HEAD / owner / 依赖：<完整 SHA、身份、依赖或无>。
+- 固定候选引用：<调度尝试中的 base/head、组件与配置；未固定写无>。
+- 窗口、最终 CI、Review 与真实合入结果：<引用调度记录及原始证据，不复制全队列>。
+
+定义见 [合入调度与 CI 分配](../standards/merge-queue.md)。本段在候选固定前整理；固定后先使用源码之外的记录保存运行结果，再按项目闭合流程回填。
+
+## 结果与验证（完成后填写）
+
+- 结果：<交付了什么；未发生写无>。
+- 验证：<结论性摘要：PASS/FAIL/SKIP/NOT_RUN 逐项；逐条证据见下表，来源标注与写入规则见 [按影响选择验证与证据](../standards/verification.md)>。
+- 影响与剩余项：<影响面、遗留风险、后续任务>。
+
+| 时间(UTC) | 命令 | 被测 HEAD | exit | 来源 | 摘要/位置 |
+|---|---|---|---|---|---|
+| <时间> | <命令；由项目 evidence 工具写入，无工具时附命令+对象 SHA+输出 digest 并标 SELF> | <HEAD> | <exit> | GATE/REPLAY/SELF | <摘要/位置> |
+
+## 回收（task-finish）
+
+- 合入核对：<分支已包含于目标分支的核对输出>。
+- 未跟踪/ignored 文件处置：<归档/可再生/存疑 逐项；无则写无>。
+- 回收动作：<worktree remove、分支删除、prune；放弃路径注明原因>。
@@ -0,0 +1,73 @@
+# Project Agent Execution Configuration
+# 项目工程 Agent 执行路由配置
+
+适用范围：定义本项目中由 AI / Agent 主导的代码实施、修复、测试与审阅所调用的 Agent CLI 路由。
+解析器识别正文中唯一一个 ```agent-execution 代码块。正文外的说明文字可根据项目需要调整。
+
+通用硬规则见 Development-Guidelines 的 [Agent 执行与协作契约](deps/development-guidelines/standards/agent-execution.md)。
+本配置为项目本地路由权威，声明具体的候选 profile、尝试次数上限、超时、集成入口与豁免路径；
+代码块须通过 `deps/development-guidelines/templates/schemas/agent-execution.schema.json` 校验后才能登记为 active。
+`fallback_to_host` 只能为 `never`；旧值 `all-unavailable` 仅供迁移工具读取，不能通过准入。
+下例的 CLI 名称、命令与路径均为占位，按项目实际执行器替换。
+
+```agent-execution
+schema: agent-execution/v1
+
+execution:
+  fresh_process_per_unit: true
+  permission: run-everything
+  fallback_to_host: never
+  max_parallel: 3
+  timeout_seconds: 7200
+  idle_timeout_seconds: 600
+  summary_bytes: 16384
+  raw_output_bytes: 10485760
+  max_attempts_per_profile: 3
+  emergency_profile: codex
+
+roles:
+  implementation: [cursor, antigravity, pi]
+  repair: [cursor, antigravity, pi]
+  test: [pi, antigravity, cursor]
+  review: [pi, antigravity, cursor, codex]
+
+profiles:
+  cursor:
+    adapter: cursor
+    executable: "cursor-agent-stable"
+    model: auto
+    # 2026-10-21 临时禁用：U3B 连续两次 fail-stop（首次 TLS 会话前故障 553B 日志、
+    # 二次 "Cannot use this model: auto"，与 Backend 配置已记录的月配额耗尽一致），
+    # 两次均未建立会话、零改动（现场保留）。恢复时改回 enabled: true。
+    enabled: false
+
+  antigravity:
+    adapter: antigravity
+    executable: agy
+    model: auto
+    # 2026-10-21 临时禁用：U3B 尝试被 400 FAILED_PRECONDITION "User location is
+    # not supported for the API use" 拒绝，0 token、未建立会话（现场保留）；
+    # 与 Backend 配置已记录的 Gemini 端点不稳一致。恢复时改回 enabled: true。
+    enabled: false
+
+  pi:
+    adapter: pi
+    executable: pi
+    model: auto
+    enabled: true
+
+  codex:
+    adapter: codex
+    executable: codex
+    model: auto
+    enabled: true
+
+integration:
+  exempt_paths: [".task/**", "docs/superpowers/specs/**", "docs/superpowers/plans/**"]
+  transport:
+    command: "scripts/agent-exec apply"
+  admission:
+    command: "scripts/agent-exec admit"
+  cleanup:
+    command: "scripts/agent-exec cleanup"
+```
@@ -34,3 +34,21 @@
 - **SSH key**：`~/.ssh/home_gitea`（ED25519，公钥指纹
  `SHA256:GYXaOXIMeloQ1fKSfdJ/ndgPQzs/Ix7Hk62ekfkfQ20`，Gitea 侧密钥名 GMKtec）；
  `~/.ssh/config` 已有 `git-ssh.u1680832.nyat.app` 主机条目，克隆/推送直接可用。
+
+## Agent 执行与新项目准入（强制）
+
+1. 缺 Guidelines 固定版本（`deps/development-guidelines` gitlink 未就绪或 HEAD 不符）时，Agent 只能只读分析，不得实施。
+2. 缺有效 `AGENT-EXECUTION.md`（`scripts/dm governance check` 未通过）时禁止 AI 实施。
+3. 缺共享执行器（`scripts/agent-exec` 或 `deps/development-manager` 不可用）时，状态为 `blocked/unavailable`，不得改由 host 替代。
+4. 不允许 host 代写实施单元；实施、修复与测试必须经 `AGENT-EXECUTION.md` 登记的候选执行。
+5. 人类直接开发按其自身授权与项目规则进行，不受本条款约束。
+6. 分支保护与规范 tag 由用户负责设置与维护，Agent 不得代为设置。
+
+入口：
+
+- 需求分流：`scripts/task triage discover|decide|verify`
+- 任务启动：`scripts/task start …`
+- PR 检查：`scripts/task pr-check …`
+- Agent 执行：`scripts/agent-exec run|apply|admit|validate`
+
+AI 实施单元只经 `scripts/agent-exec run` 按 AGENT-EXECUTION.md 运行，并经 `scripts/agent-exec apply` 集成。
@@ -12,8 +12,8 @@
 - 提交/审核/合并/发布身份与权限：仓库所有者 weichen。
 - 任务唯一来源：Gitea `weichen/Agent-Squad` 的 Issue 与 Pull Request。
 - 规范性文档目录：`policy/`。
- 过程文档仓/目录：`docs/records/`（项目内豁免，见采纳记录）。
- 交接记录位置：随任务主记录；需要落盘的交接放入 `docs/records/`。
+- 过程文档仓/目录：任务卡与交付证据在 `.task/`（模板 `.task/templates/task-card.md`）；其他过程记录在 `docs/records/`（项目内豁免，见采纳记录）。
+- 交接记录位置：随任务主记录；需要落盘的任务交接与交付证据放入 `.task/`，其他落盘交接放入 `docs/records/`。
 - 凭证查找入口及所需字段：LLM 使用环境变量 `AGENT_SQUAD_LLM_BASE_URL`、`AGENT_SQUAD_LLM_API_KEY`、`AGENT_SQUAD_LLM_MODEL`。日志采集 token 使用日志库自己的配置。文档与仓库不记录这些值。

 ## 依赖组合
@@ -45,6 +45,7 @@ git clone --recurse-submodules ssh://git@git-ssh.u1680832.nyat.app:31411/weichen
 | 模块/接口 | 仓库根：`GOWORK=off go vet ./... && GOWORK=off go test -count=1 ./...` | 需要已初始化的 Agent-Native 子模块；默认 stub，不调用真实模型 |
 | 数据/权限/恢复 | conformance 与 store 测试覆盖运行落盘、HITL 恢复 | 无共享数据库 |
 | UI/执行后端 | `GOWORK=off go test -count=1 ./server/... ./conformance/...` | 不覆盖浏览器；SSE 由测试客户端验证 |
+| 治理与 Agent 执行配置 | 仓库根：`scripts/agent-exec validate --repo .` 与 `scripts/dm governance check --repo .` | 需 `deps/development-manager` 已按 gitlink 检出（首次 `dm` 调用会构建 dm-server）；只核验 AGENT-EXECUTION.md 与治理文件，不证明实施单元正确 |

 ## 环境与数据

@@ -0,0 +1,179 @@
+// Package a2a 把小队接入 A2A（Agent2Agent）1.0 互通面（Q3 WP4）：
+//   - 发布侧：BuildCard 从角色注册表与蓝图注册表渲染 Agent Card（描述本
+//     团队能力），CardHandler 以 JSON 服务（挂载在
+//     /.well-known/agent-card.json）；
+//   - 消费侧：FetchCard 发现远端 Agent Card（fail-closed 校验必备字段），
+//     Executor 实现 squad.Executor——message/send 发起任务，tasks/get 轮询
+//     A2A 任务状态机（submitted/working/input-required → completed/failed/
+//     canceled），终态映射 ExecResult：completed 的 artifacts（TextPart/
+//     FilePart）落成 patches/目标文件；failed/canceled/input-required 一律
+//     Success=false（跨进程挂起不在本适配面内，fail-closed 交引擎重试/降级
+//     语义处理）。
+//
+// 传输为 JSON-RPC 2.0 over HTTP（A2A 1.0 形态）；不引入协议之外的依赖。
+package a2a
+
+import (
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"strings"
+
+	"agent-squad/squad"
+)
+
+// ProtocolVersion 对齐的 A2A 协议版本。
+const ProtocolVersion = "1.0"
+
+// WellKnownPath 是 Agent Card 的发现路径（A2A 1.0）。
+const WellKnownPath = "/.well-known/agent-card.json"
+
+// AgentCard 是 A2A 1.0 Agent Card（发布与消费共用形态）。
+type AgentCard struct {
+	ProtocolVersion string `json:"protocolVersion"`
+	Name            string `json:"name"`
+	Description     string `json:"description,omitempty"`
+	// URL 是 JSON-RPC 服务端点（消费侧据此发 message/send）。
+	URL         string   `json:"url"`
+	Version     string   `json:"version"`
+	Capabilities CardCapabilities `json:"capabilities"`
+	DefaultInputModes  []string `json:"defaultInputModes,omitempty"`
+	DefaultOutputModes []string `json:"defaultOutputModes,omitempty"`
+	Skills      []AgentSkill `json:"skills,omitempty"`
+}
+
+// CardCapabilities 能力声明（本适配面：非流式、无推送、无状态历史）。
+type CardCapabilities struct {
+	Streaming              bool `json:"streaming"`
+	PushNotifications      bool `json:"pushNotifications"`
+	StateTransitionHistory bool `json:"stateTransitionHistory"`
+}
+
+// AgentSkill 是一项可被远端调用的能力。
+type AgentSkill struct {
+	ID          string   `json:"id"`
+	Name        string   `json:"name"`
+	Description string   `json:"description,omitempty"`
+	Tags        []string `json:"tags,omitempty"`
+	Examples    []string `json:"examples,omitempty"`
+}
+
+// CardOptions 覆盖 BuildCard 的身份字段（空则用缺省）。
+type CardOptions struct {
+	Name        string
+	Description string
+	// URL JSON-RPC 端点（必填：远端没有端点就无法调用本团队）。
+	URL string
+	// Version 卡片版本（空 = "1.0.0"）。
+	Version string
+	// Skills 显式能力清单；空 = 从注册表派生。
+	Skills []AgentSkill
+}
+
+// BuildCard 从角色注册表与蓝图注册表渲染 Agent Card（描述本团队能力）。
+// 角色 → 技能（关键词作 tags）；蓝图 → 技能（match 词作 examples）。
+func BuildCard(reg *squad.RoleRegistry, bps *squad.BlueprintRegistry, opts CardOptions) (AgentCard, error) {
+	if strings.TrimSpace(opts.URL) == "" {
+		return AgentCard{}, fmt.Errorf("a2a: Agent Card 需要 JSON-RPC 端点 URL")
+	}
+	card := AgentCard{
+		ProtocolVersion: ProtocolVersion,
+		Name:            orDefault(opts.Name, "agent-squad"),
+		Description:     opts.Description,
+		URL:             opts.URL,
+		Version:         orDefault(opts.Version, "1.0.0"),
+		DefaultInputModes:  []string{"text/plain"},
+		DefaultOutputModes: []string{"text/plain"},
+		Skills:          append([]AgentSkill(nil), opts.Skills...),
+	}
+	if len(card.Skills) == 0 {
+		if reg != nil {
+			for _, role := range reg.List() {
+				desc := firstLine(role.Instructions)
+				if desc == "" {
+					desc = role.Name
+				}
+				card.Skills = append(card.Skills, AgentSkill{
+					ID: role.Role, Name: orDefault(role.Name, role.Role),
+					Description: desc, Tags: append([]string(nil), role.Keywords...),
+				})
+			}
+		}
+		if bps != nil {
+			for _, bp := range bps.List() {
+				card.Skills = append(card.Skills, AgentSkill{
+					ID: bp.Key, Name: orDefault(bp.Title, bp.Key),
+					Description: bp.Description, Examples: append([]string(nil), bp.Match...),
+				})
+			}
+		}
+	}
+	return card, nil
+}
+
+// CardHandler 以 JSON 服务 Agent Card（调用方挂载到 WellKnownPath）。
+func CardHandler(card AgentCard) http.Handler {
+	body, _ := json.Marshal(card)
+	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write(body)
+	})
+}
+
+// FetchCard 发现并校验远端 Agent Card（fail-closed：必备字段缺失即拒）。
+func FetchCard(client *http.Client, wellKnownURL string) (AgentCard, error) {
+	if client == nil {
+		client = http.DefaultClient
+	}
+	resp, err := client.Get(wellKnownURL)
+	if err != nil {
+		return AgentCard{}, fmt.Errorf("a2a: 获取 Agent Card: %w", err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		return AgentCard{}, fmt.Errorf("a2a: Agent Card 端点返回 %d", resp.StatusCode)
+	}
+	var card AgentCard
+	if err := json.NewDecoder(resp.Body).Decode(&card); err != nil {
+		return AgentCard{}, fmt.Errorf("a2a: Agent Card 解析失败: %w", err)
+	}
+	if err := validateCard(card); err != nil {
+		return AgentCard{}, err
+	}
+	return card, nil
+}
+
+// validateCard 消费侧最小校验（A2A 1.0 必备字段）。
+func validateCard(card AgentCard) error {
+	if card.ProtocolVersion == "" {
+		return fmt.Errorf("a2a: Agent Card 缺 protocolVersion")
+	}
+	if strings.TrimSpace(card.Name) == "" {
+		return fmt.Errorf("a2a: Agent Card 缺 name")
+	}
+	if strings.TrimSpace(card.URL) == "" {
+		return fmt.Errorf("a2a: Agent Card 缺 url")
+	}
+	if strings.TrimSpace(card.Version) == "" {
+		return fmt.Errorf("a2a: Agent Card 缺 version")
+	}
+	return nil
+}
+
+func orDefault(v, def string) string {
+	if strings.TrimSpace(v) != "" {
+		return v
+	}
+	return def
+}
+
+func firstLine(s string) string {
+	s = strings.TrimSpace(s)
+	if i := strings.IndexByte(s, '\n'); i >= 0 {
+		s = s[:i]
+	}
+	if len(s) > 120 {
+		s = s[:120] + "…"
+	}
+	return s
+}
@@ -0,0 +1,345 @@
+package a2a
+
+import (
+	"context"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"agent-squad/squad"
+)
+
+// conformance_test.go — Q3 WP4 验收：双向 conformance（本地假 A2A 服务端）。
+// 发布侧：BuildCard + CardHandler 在 well-known 路径服务合法 Agent Card；
+// 消费侧：FetchCard 发现校验 + Executor 走完任务状态机
+// （submitted→working→completed / failed / input-required），终态对齐
+// squad.ExecResult。
+
+// fakeA2AServer 最小 A2A 1.0 服务端：well-known 卡片 + JSON-RPC
+// （message/send 建任务；tasks/get 按脚本推进状态）。
+type fakeA2AServer struct {
+	mu     sync.Mutex
+	card   AgentCard
+	tasks  map[string]*Task
+	plan   []string // 每次 tasks/get 推进到的状态序列（走完停在末态）
+	calls  map[string]int
+	artifacts []Artifact
+	statusMsg string
+}
+
+func newFakeServer(t *testing.T, card AgentCard, plan []string, artifacts []Artifact) *fakeA2AServer {
+	t.Helper()
+	s := &fakeA2AServer{card: card, tasks: map[string]*Task{}, plan: plan,
+		calls: map[string]int{}, artifacts: artifacts}
+	return s
+}
+
+func (s *fakeA2AServer) handler() http.Handler {
+	mux := http.NewServeMux()
+	mux.HandleFunc(WellKnownPath, func(w http.ResponseWriter, r *http.Request) {
+		s.mu.Lock()
+		body, _ := json.Marshal(s.card)
+		s.mu.Unlock()
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write(body)
+	})
+	mux.HandleFunc("/rpc", func(w http.ResponseWriter, r *http.Request) {
+		var req rpcRequest
+		if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+			writeRPCError(w, 0, -32700, "parse error")
+			return
+		}
+		s.mu.Lock()
+		defer s.mu.Unlock()
+		s.calls[req.Method]++
+		switch req.Method {
+		case "message/send":
+			id := fmt.Sprintf("task_%d", len(s.tasks)+1)
+			state := StateSubmitted
+			if len(s.plan) > 0 {
+				state = s.plan[0]
+			}
+			task := &Task{Kind: "task", ID: id, ContextID: "ctx_1",
+				Status: TaskStatus{State: state, Message: s.statusMessage()}}
+			s.tasks[id] = task
+			writeRPCResult(w, req.ID, task)
+		case "tasks/get":
+			var params struct {
+				ID string `json:"id"`
+			}
+			raw, _ := json.Marshal(req.Params)
+			_ = json.Unmarshal(raw, &params)
+			task, ok := s.tasks[params.ID]
+			if !ok {
+				writeRPCError(w, req.ID, -32001, "task not found")
+				return
+			}
+			// 按脚本推进状态机（停留在终态）。
+			if !isTerminal(task.Status.State) && len(s.plan) > 0 {
+				n := s.calls["tasks/get#"+params.ID]
+				s.calls["tasks/get#"+params.ID] = n + 1
+				idx := n + 1
+				if idx >= len(s.plan) {
+					idx = len(s.plan) - 1
+				}
+				task.Status.State = s.plan[idx]
+				task.Status.Message = s.statusMessage()
+				if task.Status.State == StateCompleted {
+					task.Artifacts = s.artifacts
+				}
+			}
+			writeRPCResult(w, req.ID, task)
+		default:
+			writeRPCError(w, req.ID, -32601, "method not found")
+		}
+	})
+	return mux
+}
+
+func (s *fakeA2AServer) statusMessage() *Message {
+	if s.statusMsg == "" {
+		return nil
+	}
+	return &Message{Kind: "message", Role: "agent",
+		Parts: []Part{{Kind: "text", Text: s.statusMsg}}}
+}
+
+func writeRPCResult(w http.ResponseWriter, id int64, result any) {
+	body, _ := json.Marshal(result)
+	env, _ := json.Marshal(rpcResponse{JSONRPC: "2.0", ID: id, Result: body})
+	w.Header().Set("Content-Type", "application/json")
+	_, _ = w.Write(env)
+}
+
+func writeRPCError(w http.ResponseWriter, id int64, code int, msg string) {
+	env, _ := json.Marshal(rpcResponse{JSONRPC: "2.0", ID: id, Error: &rpcError{Code: code, Message: msg}})
+	w.Header().Set("Content-Type", "application/json")
+	_, _ = w.Write(env)
+}
+
+// ---- 发布侧 ----
+
+func TestBuildCardFromRegistries(t *testing.T) {
+	reg := squad.NewRoleRegistry()
+	for _, r := range []squad.RoleSpec{
+		{Role: "researcher", Name: "调研员", Keywords: []string{"调研", "检索"},
+			Instructions: "围绕目标调研关键问题。\n第二行不进摘要。"},
+		{Role: "writer", Name: "撰稿员", Keywords: []string{"撰写"}},
+	} {
+		if err := reg.Register(r); err != nil {
+			t.Fatal(err)
+		}
+	}
+	bps := squad.NewBlueprintRegistry()
+	if err := bps.Register(squad.Blueprint{Key: "brief", Title: "简报",
+		Description: "生成简报", Match: []string{"简报", "brief"},
+		Steps: []squad.StepSpec{{ID: "s1", Role: "writer", Title: "成文"}}}); err != nil {
+		t.Fatal(err)
+	}
+	card, err := BuildCard(reg, bps, CardOptions{
+		Name: "示例小队", Description: "演示团队能力", URL: "https://team.example/rpc",
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if card.ProtocolVersion != ProtocolVersion || card.Version != "1.0.0" {
+		t.Fatalf("card = %+v", card)
+	}
+	byID := map[string]AgentSkill{}
+	for _, s := range card.Skills {
+		byID[s.ID] = s
+	}
+	if byID["researcher"].Name != "调研员" || byID["researcher"].Description != "围绕目标调研关键问题。" {
+		t.Fatalf("role skill = %+v", byID["researcher"])
+	}
+	if len(byID["researcher"].Tags) != 2 {
+		t.Fatalf("tags = %v", byID["researcher"].Tags)
+	}
+	if byID["brief"].Name != "简报" || strings.Join(byID["brief"].Examples, ",") != "简报,brief" {
+		t.Fatalf("blueprint skill = %+v", byID["brief"])
+	}
+	if _, err := BuildCard(reg, bps, CardOptions{}); err == nil {
+		t.Fatal("缺 URL 必须 fail-closed")
+	}
+}
+
+// CardHandler 以 JSON 服务卡片（发布端点形态）。
+func TestCardHandlerServesJSON(t *testing.T) {
+	card := AgentCard{ProtocolVersion: ProtocolVersion, Name: "小队", URL: "https://x/rpc", Version: "1.0.0"}
+	ts := httptest.NewServer(CardHandler(card))
+	defer ts.Close()
+	resp, err := http.Get(ts.URL + WellKnownPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer resp.Body.Close()
+	if ct := resp.Header.Get("Content-Type"); ct != "application/json" {
+		t.Fatalf("content-type = %s", ct)
+	}
+	var got AgentCard
+	if err := json.NewDecoder(resp.Body).Decode(&got); err != nil {
+		t.Fatal(err)
+	}
+	if got.Name != "小队" || got.ProtocolVersion != ProtocolVersion {
+		t.Fatalf("card = %+v", got)
+	}
+}
+
+// ---- 消费侧：发现 + 状态机 ----
+
+func TestFetchCardAndExecuteCompleted(t *testing.T) {
+	fileBytes := base64.StdEncoding.EncodeToString([]byte("notes 内容"))
+	artifacts := []Artifact{{
+		ArtifactID: "art1", Name: "调研结果",
+		Parts: []Part{
+			{Kind: "text", Text: "# 调研结论\n\n结论正文，长度足够作为交付物内容。"},
+			{Kind: "file", File: &FilePart{Name: "notes.md", Bytes: fileBytes, MimeType: "text/markdown"}},
+		},
+	}}
+	card := AgentCard{ProtocolVersion: ProtocolVersion, Name: "远端小队",
+		URL: "PLACEHOLDER", Version: "1.0.0"}
+	srv := newFakeServer(t, card, []string{StateWorking, StateWorking, StateCompleted}, artifacts)
+	ts := httptest.NewServer(srv.handler())
+	defer ts.Close()
+	srv.mu.Lock()
+	srv.card.URL = ts.URL + "/rpc"
+	srv.mu.Unlock()
+
+	// 发现：well-known 路径取卡并校验。
+	got, err := FetchCard(ts.Client(), ts.URL+WellKnownPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got.Name != "远端小队" || got.URL != ts.URL+"/rpc" {
+		t.Fatalf("card = %+v", got)
+	}
+
+	// 执行：submitted/working → completed，工件映射 patches。
+	var states []string
+	ex := NewExecutor(got)
+	ex.Client = ts.Client()
+	ex.PollInterval = 5 * time.Millisecond
+	ex.OnState = func(s string) { states = append(states, s) }
+	res, err := ex.Execute(context.Background(), squad.ExecRequest{
+		Prompt: "调研主题 X", Targets: []string{"output.md"},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !res.Success {
+		t.Fatalf("result = %+v", res)
+	}
+	if string(res.Patches["notes.md"]) != "notes 内容" {
+		t.Fatalf("file part 未落成 patch：%q", res.Patches["notes.md"])
+	}
+	if !strings.Contains(string(res.Patches["output.md"]), "调研结论") {
+		t.Fatalf("文本部件未写入目标：%q", res.Patches["output.md"])
+	}
+	joined := strings.Join(states, ",")
+	if !strings.Contains(joined, StateWorking) || !strings.HasSuffix(joined, StateCompleted) {
+		t.Fatalf("状态机迁移 = %v", states)
+	}
+	srv.mu.Lock()
+	sends := srv.calls["message/send"]
+	srv.mu.Unlock()
+	if sends != 1 {
+		t.Fatalf("message/send = %d", sends)
+	}
+}
+
+func TestExecuteMapsFailedAndInputRequired(t *testing.T) {
+	card := AgentCard{ProtocolVersion: ProtocolVersion, Name: "远端", URL: "PLACEHOLDER", Version: "1"}
+	srv := newFakeServer(t, card, []string{StateWorking, StateFailed}, nil)
+	srv.statusMsg = "远端执行崩溃"
+	ts := httptest.NewServer(srv.handler())
+	defer ts.Close()
+	srv.mu.Lock()
+	srv.card.URL = ts.URL + "/rpc"
+	srv.mu.Unlock()
+
+	ex := NewExecutor(card)
+	ex.BaseURL = ts.URL + "/rpc"
+	ex.Client = ts.Client()
+	ex.PollInterval = 5 * time.Millisecond
+	res, err := ex.Execute(context.Background(), squad.ExecRequest{Prompt: "x", Targets: []string{"output.md"}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res.Success || !strings.Contains(res.ErrorReason, "failed") || !strings.Contains(res.ErrorReason, "远端执行崩溃") {
+		t.Fatalf("failed 映射 = %+v", res)
+	}
+
+	// input-required：fail-closed（跨进程人工输入不在适配面内）。
+	srv2 := newFakeServer(t, card, []string{StateInputRequired}, nil)
+	srv2.statusMsg = "请补充授权"
+	ts2 := httptest.NewServer(srv2.handler())
+	defer ts2.Close()
+	ex2 := &Executor{BaseURL: ts2.URL + "/rpc", Client: ts2.Client(), PollInterval: 5 * time.Millisecond}
+	res2, err := ex2.Execute(context.Background(), squad.ExecRequest{Prompt: "x", Targets: []string{"output.md"}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res2.Success || !strings.Contains(res2.ErrorReason, "input-required") {
+		t.Fatalf("input-required 映射 = %+v", res2)
+	}
+}
+
+// 直接回 Message（无任务态）→ 文本落目标。
+func TestExecuteDirectMessageResult(t *testing.T) {
+	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		var req rpcRequest
+		_ = json.NewDecoder(r.Body).Decode(&req)
+		writeRPCResult(w, req.ID, Message{Kind: "message", Role: "agent",
+			Parts: []Part{{Kind: "text", Text: "直接回答正文"}}})
+	}))
+	defer ts.Close()
+	ex := &Executor{BaseURL: ts.URL, Client: ts.Client()}
+	res, err := ex.Execute(context.Background(), squad.ExecRequest{Prompt: "x", Targets: []string{"output.md"}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !res.Success || !strings.Contains(string(res.Patches["output.md"]), "直接回答正文") {
+		t.Fatalf("res = %+v", res)
+	}
+}
+
+// 传输/协议错误 → error（引擎按瞬态评估），不吞成成功。
+func TestExecuteProtocolErrors(t *testing.T) {
+	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		writeRPCError(w, 1, -32601, "method not found")
+	}))
+	defer ts.Close()
+	ex := &Executor{BaseURL: ts.URL, Client: ts.Client()}
+	if _, err := ex.Execute(context.Background(), squad.ExecRequest{Prompt: "x"}); err == nil ||
+		!strings.Contains(err.Error(), "协议错误") {
+		t.Fatalf("err = %v", err)
+	}
+	// ctx 取消传播。
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+	slow := newFakeServer(t, AgentCard{ProtocolVersion: "1.0", Name: "x", URL: ts.URL, Version: "1"},
+		[]string{StateWorking}, nil)
+	ts2 := httptest.NewServer(slow.handler())
+	defer ts2.Close()
+	ex2 := &Executor{BaseURL: ts2.URL + "/rpc", Client: ts2.Client(), PollInterval: time.Second}
+	if _, err := ex2.Execute(ctx, squad.ExecRequest{Prompt: "x"}); err == nil {
+		t.Fatal("取消必须传播")
+	}
+}
+
+// FetchCard fail-closed：缺必备字段拒绝。
+func TestFetchCardValidates(t *testing.T) {
+	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_, _ = w.Write([]byte(`{"protocolVersion":"1.0","name":"","url":"","version":""}`))
+	}))
+	defer ts.Close()
+	if _, err := FetchCard(ts.Client(), ts.URL); err == nil || !strings.Contains(err.Error(), "name") {
+		t.Fatalf("err = %v", err)
+	}
+}
@@ -0,0 +1,331 @@
+package a2a
+
+import (
+	"bytes"
+	"context"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"sync/atomic"
+	"time"
+
+	"agent-squad/squad"
+)
+
+// executor.go — 远端智能体 Executor 客户端（A2A 任务状态机对齐 ExecResult）。
+
+// 任务状态（A2A 1.0 TaskState）。
+const (
+	StateSubmitted    = "submitted"
+	StateWorking      = "working"
+	StateInputRequired = "input-required"
+	StateCompleted    = "completed"
+	StateCanceled     = "canceled"
+	StateFailed       = "failed"
+	StateRejected     = "rejected"
+)
+
+// DefaultPollInterval 是 tasks/get 轮询缺省间隔。
+const DefaultPollInterval = 500 * time.Millisecond
+
+// Executor 把远端 A2A 智能体接成 squad.Executor：一次步骤执行 = 一个
+// A2A 任务（message/send 发起，tasks/get 轮询到终态）。
+type Executor struct {
+	// BaseURL 远端 JSON-RPC 端点（Agent Card 的 url）。
+	BaseURL string
+	// Client HTTP 客户端（nil = http.DefaultClient）。
+	Client *http.Client
+	// PollInterval 轮询间隔（0 = DefaultPollInterval）。
+	PollInterval time.Duration
+	// OnState 可选状态回调（观测任务状态机迁移）。
+	OnState func(state string)
+
+	callID atomic.Int64
+}
+
+// NewExecutor 按 Agent Card 构造远端执行器。
+func NewExecutor(card AgentCard) *Executor {
+	return &Executor{BaseURL: card.URL}
+}
+
+// Name 实现 squad.Executor。
+func (e *Executor) Name() string { return "a2a" }
+
+// ---- A2A 1.0 报文形态（JSON-RPC 2.0） ----
+
+type rpcRequest struct {
+	JSONRPC string `json:"jsonrpc"`
+	ID      int64  `json:"id"`
+	Method  string `json:"method"`
+	Params  any    `json:"params"`
+}
+
+type rpcError struct {
+	Code    int    `json:"code"`
+	Message string `json:"message"`
+}
+
+type rpcResponse struct {
+	JSONRPC string          `json:"jsonrpc"`
+	ID      int64           `json:"id"`
+	Result  json.RawMessage `json:"result,omitempty"`
+	Error   *rpcError       `json:"error,omitempty"`
+}
+
+// Part 是消息/工件的部件（text | file | data）。
+type Part struct {
+	Kind string     `json:"kind"`
+	Text string     `json:"text,omitempty"`
+	File *FilePart  `json:"file,omitempty"`
+	Data *json.RawMessage `json:"data,omitempty"`
+}
+
+// FilePart 文件部件：bytes（base64）或 uri 二选一。
+type FilePart struct {
+	Name     string `json:"name"`
+	Bytes    string `json:"bytes,omitempty"`
+	URI      string `json:"uri,omitempty"`
+	MimeType string `json:"mimeType,omitempty"`
+}
+
+// Message 是一条 A2A 消息。
+type Message struct {
+	Kind      string `json:"kind"`
+	Role      string `json:"role"`
+	Parts     []Part `json:"parts"`
+	MessageID string `json:"messageId"`
+}
+
+// TaskStatus 任务状态。
+type TaskStatus struct {
+	State    string   `json:"state"`
+	Message  *Message `json:"message,omitempty"`
+	Timestamp string  `json:"timestamp,omitempty"`
+}
+
+// Artifact 任务工件。
+type Artifact struct {
+	ArtifactID string `json:"artifactId,omitempty"`
+	Name       string `json:"name,omitempty"`
+	Parts      []Part `json:"parts"`
+}
+
+// Task 是 A2A 任务（状态机载体）。
+type Task struct {
+	Kind      string     `json:"kind"`
+	ID        string     `json:"id"`
+	ContextID string     `json:"contextId,omitempty"`
+	Status    TaskStatus `json:"status"`
+	Artifacts []Artifact `json:"artifacts,omitempty"`
+}
+
+// Execute 实现 squad.Executor：发起任务并轮询到终态，映射 ExecResult。
+// 传输/协议错误返回 error（引擎按瞬态失败评估）；任务级失败
+// （failed/canceled/rejected/input-required）返回 Success=false 的结果
+// （fail-closed：跨进程人工输入不在本适配面内）。
+func (e *Executor) Execute(ctx context.Context, req squad.ExecRequest) (*squad.ExecResult, error) {
+	if strings.TrimSpace(e.BaseURL) == "" {
+		return nil, fmt.Errorf("a2a: executor 缺 BaseURL")
+	}
+	prompt := strings.TrimSpace(req.Prompt)
+	if prompt == "" {
+		prompt = strings.TrimSpace(req.Instructions)
+	}
+	msg := Message{
+		Kind: "message", Role: "user",
+		Parts:     []Part{{Kind: "text", Text: prompt}},
+		MessageID: fmt.Sprintf("msg-%d-%d", time.Now().UnixNano(), e.callID.Add(1)),
+	}
+	raw, err := e.call(ctx, "message/send", map[string]any{"message": msg})
+	if err != nil {
+		return nil, err
+	}
+	// 服务端可直接回 Message（无任务态）或 Task。
+	var probe struct {
+		Kind string `json:"kind"`
+	}
+	_ = json.Unmarshal(raw, &probe)
+	if probe.Kind == "message" {
+		var m Message
+		if err := json.Unmarshal(raw, &m); err != nil {
+			return nil, fmt.Errorf("a2a: message 结果解析失败: %w", err)
+		}
+		return e.resultFromParts(m.Parts, req.Targets), nil
+	}
+	var task Task
+	if err := json.Unmarshal(raw, &task); err != nil {
+		return nil, fmt.Errorf("a2a: task 结果解析失败: %w", err)
+	}
+	e.observe(task.Status.State)
+	if shouldStop(task.Status.State) {
+		return e.resultFromTask(task, req.Targets), nil
+	}
+	// 轮询任务状态机直至终态（ctx 取消传播）。
+	interval := e.PollInterval
+	if interval <= 0 {
+		interval = DefaultPollInterval
+	}
+	ticker := time.NewTicker(interval)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-ctx.Done():
+			return nil, ctx.Err()
+		case <-ticker.C:
+		}
+		raw, err := e.call(ctx, "tasks/get", map[string]any{"id": task.ID})
+		if err != nil {
+			return nil, err
+		}
+		var next Task
+		if err := json.Unmarshal(raw, &next); err != nil {
+			return nil, fmt.Errorf("a2a: tasks/get 解析失败: %w", err)
+		}
+		if next.Status.State != task.Status.State {
+			task = next
+			e.observe(task.Status.State)
+		} else {
+			task = next
+		}
+		if shouldStop(task.Status.State) {
+			return e.resultFromTask(task, req.Targets), nil
+		}
+	}
+}
+
+// call 发一次 JSON-RPC 调用并取 result。
+func (e *Executor) call(ctx context.Context, method string, params any) (json.RawMessage, error) {
+	client := e.Client
+	if client == nil {
+		client = http.DefaultClient
+	}
+	body, _ := json.Marshal(rpcRequest{
+		JSONRPC: "2.0", ID: e.callID.Add(1), Method: method, Params: params,
+	})
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, e.BaseURL, bytes.NewReader(body))
+	if err != nil {
+		return nil, err
+	}
+	req.Header.Set("Content-Type", "application/json")
+	resp, err := client.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("a2a: %s 传输失败: %w", method, err)
+	}
+	defer resp.Body.Close()
+	data, err := io.ReadAll(io.LimitReader(resp.Body, 8<<20))
+	if err != nil {
+		return nil, fmt.Errorf("a2a: %s 读取响应失败: %w", method, err)
+	}
+	if resp.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("a2a: %s 返回 HTTP %d", method, resp.StatusCode)
+	}
+	var env rpcResponse
+	if err := json.Unmarshal(data, &env); err != nil {
+		return nil, fmt.Errorf("a2a: %s 响应不是合法 JSON-RPC: %w", method, err)
+	}
+	if env.Error != nil {
+		return nil, fmt.Errorf("a2a: %s 协议错误 %d: %s", method, env.Error.Code, env.Error.Message)
+	}
+	return env.Result, nil
+}
+
+func (e *Executor) observe(state string) {
+	if e.OnState != nil {
+		e.OnState(state)
+	}
+}
+
+func isTerminal(state string) bool {
+	switch state {
+	case StateCompleted, StateFailed, StateCanceled, StateRejected:
+		return true
+	}
+	return false
+}
+
+// shouldStop 轮询停止条件：真终态，或 input-required（跨进程人工输入不在
+// 本适配面内，fail-closed 交引擎语义处理，不死等）。
+func shouldStop(state string) bool {
+	return isTerminal(state) || state == StateInputRequired
+}
+
+// resultFromTask 终态任务 → ExecResult。
+func (e *Executor) resultFromTask(task Task, targets []string) *squad.ExecResult {
+	switch task.Status.State {
+	case StateCompleted:
+		var parts []Part
+		for _, art := range task.Artifacts {
+			parts = append(parts, art.Parts...)
+		}
+		return e.resultFromParts(parts, targets)
+	case StateInputRequired:
+		return &squad.ExecResult{Success: false,
+			ErrorReason: "a2a: 远端要求人工输入（input-required）：" + statusText(task.Status)}
+	case StateCanceled:
+		return &squad.ExecResult{Success: false,
+			ErrorReason: "a2a: 远端任务被取消（canceled）：" + statusText(task.Status)}
+	case StateRejected:
+		return &squad.ExecResult{Success: false,
+			ErrorReason: "a2a: 远端拒绝任务（rejected）：" + statusText(task.Status)}
+	default: // failed 及未知终态一律 fail-closed
+		return &squad.ExecResult{Success: false,
+			ErrorReason: "a2a: 远端任务失败（" + task.Status.State + "）：" + statusText(task.Status)}
+	}
+}
+
+// resultFromParts 部件流 → patches：FilePart(bytes) 按名落 patch；
+// 文本部件拼接后写入全部声明目标（无文件工件时的交付物兜底，
+// 与其他后端「正文写目标文件」契约一致）。
+func (e *Executor) resultFromParts(parts []Part, targets []string) *squad.ExecResult {
+	patches := map[string][]byte{}
+	var text strings.Builder
+	for _, p := range parts {
+		switch p.Kind {
+		case "text":
+			if text.Len() > 0 {
+				text.WriteString("\n")
+			}
+			text.WriteString(p.Text)
+		case "file":
+			if p.File == nil || p.File.Name == "" {
+				continue
+			}
+			data, err := base64.StdEncoding.DecodeString(p.File.Bytes)
+			if err != nil {
+				continue // uri 形态与坏 base64 不落 patch（消费侧最小面）
+			}
+			patches[p.File.Name] = data
+		}
+	}
+	body := strings.TrimSpace(text.String())
+	if len(targets) == 0 {
+		targets = []string{"output.md"}
+	}
+	if body != "" {
+		for _, t := range targets {
+			if _, ok := patches[t]; !ok {
+				patches[t] = []byte(body + "\n")
+			}
+		}
+	}
+	if len(patches) == 0 {
+		return &squad.ExecResult{Success: false, ErrorReason: "a2a: 完成任务无任何工件"}
+	}
+	return &squad.ExecResult{Success: true, Patches: patches, Logs: body}
+}
+
+func statusText(st TaskStatus) string {
+	if st.Message == nil {
+		return ""
+	}
+	var b strings.Builder
+	for _, p := range st.Message.Parts {
+		if p.Kind == "text" {
+			b.WriteString(p.Text)
+		}
+	}
+	return b.String()
+}
@@ -0,0 +1,11 @@
+module agent-squad/adapters/a2a
+
+go 1.27
+
+require agent-squad v0.0.0
+
+require error-code-system v0.0.0 // indirect
+
+replace agent-squad => ../..
+
+replace error-code-system => ../../deps/error-code-system
@@ -3,7 +3,11 @@ package native

 import (
 	"context"
+	"crypto/sha256"
+	"encoding/hex"
+	"encoding/json"
 	"fmt"
+	"reflect"
 	"strings"
 	"time"

@@ -52,12 +56,58 @@ type Config struct {
 	PermissionAsk func(ctx context.Context, call llm.ToolCall) runtime.AskVerdict
 }

+// Identity 返回塑造 Executor 的配置摘要（模型/客户端/工具/Exec 等）。
+// RegisterFull 经 BackendResolver.Register 替换工厂时会失效缓存；角色侧
+// BashPermission 等由 squad.ConfigIdentity 纳入 Resolve 缓存键，故 allow→deny
+// 在下一次 Resolve 即生效，无需重启进程。
+func (c Config) Identity() string {
+	var b strings.Builder
+	b.WriteString("model=")
+	b.WriteString(c.DefaultModel)
+	b.WriteString("\x00turns=")
+	b.WriteString(fmt.Sprintf("%d", c.MaxTurns))
+	b.WriteString("\x00timeout=")
+	b.WriteString(c.Timeout.String())
+	if c.Client != nil {
+		b.WriteString("\x00base=")
+		b.WriteString(c.Client.BaseURL)
+		b.WriteString("\x00key=")
+		b.WriteString(c.Client.APIKey)
+	}
+	if c.Exec != nil {
+		b.WriteString("\x00exec.bash=")
+		b.WriteString(c.Exec.Bash)
+		b.WriteString("\x00exec.python=")
+		b.WriteString(c.Exec.Python)
+	}
+	if c.Search != nil {
+		b.WriteString("\x00search=1")
+	}
+	if c.Fetch != nil {
+		b.WriteString("\x00fetch=1")
+	}
+	if c.MCP != nil {
+		b.WriteString("\x00mcp=1")
+	}
+	if len(c.SkillsRoots) > 0 {
+		b.WriteString("\x00skills=")
+		b.WriteString(strings.Join(c.SkillsRoots, ","))
+	}
+	if c.CodeOrchestration {
+		b.WriteString("\x00code=1")
+	}
+	sum := sha256.Sum256([]byte(b.String()))
+	return hex.EncodeToString(sum[:16])
+}
+
 // Register 把 "agent-native" 工厂注册进解析器。
 func Register(r *squad.BackendResolver, client *llm.Client, defaultModel string, maxTurns int, timeout time.Duration) {
 	RegisterFull(r, Config{Client: client, DefaultModel: defaultModel, MaxTurns: maxTurns, Timeout: timeout})
 }

 // RegisterFull 注册带工具装配的 "agent-native" 工厂。
+// 再次调用会替换工厂并失效该后端缓存（见 BackendResolver.Register）；
+// 配置摘要见 Config.Identity，角色侧见 squad.ConfigIdentity。
 func RegisterFull(r *squad.BackendResolver, cfg Config) {
 	r.Register("agent-native", func(role squad.RoleSpec) (squad.Executor, error) {
 		opts, err := buildOptions(cfg, role)
@@ -147,10 +197,62 @@ func toNative(req squad.ExecRequest) exec.Request {
 }

 func fromNative(res *exec.Result) *squad.ExecResult {
-	return &squad.ExecResult{
-		Success: res.Success, Patches: res.Patches, Logs: res.Logs, ErrorReason: res.ErrorReason,
-		Retryable: res.Retryable, PromptTokens: res.PromptTokens, CompletionTokens: res.CompletionTokens,
+	if res == nil {
+		return nil
 	}
+	out := &squad.ExecResult{
+		Success:          res.Success,
+		Patches:          res.Patches,
+		Logs:             res.Logs,
+		ErrorReason:      res.ErrorReason,
+		Retryable:        res.Retryable,
+		PromptTokens:     res.PromptTokens,
+		CompletionTokens: res.CompletionTokens,
+	}
+	// F05/U3b（Squad 侧 contract）：
+	//   - native exec.Result 在后续 release 会新增 SideEffects、Canceled、Aborted 字段；
+	//     本地 deps gitlink 若尚未跟进，从 JSON 探测拿不到这些字段时回落到
+	//     「legacy reading: side effects = possible」，与既有调用方旧行为一致。
+	//   - 在不影响 ExecResult 默认值的前提下，用 JSON 探测而非硬性修改 deps。
+	probe, err := probeNativeResultExtras(res)
+	if err == nil {
+		out.SideEffects = probe.SideEffects
+		out.Canceled = probe.Canceled
+		out.Aborted = probe.Aborted
+	}
+	return out
+}
+
+// nativeResultExtras 探测 native exec.Result 上的可选字段（结构化 JSON）。
+// 缺省（字段不存在或为零值时）按 legacy reading：SideEffects 空串视为 "possible"；
+// 调用方侧 ExecResult.IsRetryableSideEffectReports() 已经处理该回退。
+type nativeResultExtras struct {
+	SideEffects squad.ExecSideEffects `json:"side_effects,omitempty"`
+	Canceled    bool                  `json:"canceled,omitempty"`
+	Aborted     bool                  `json:"aborted,omitempty"`
+}
+
+// probeNativeResultExtras 通过 JSON 探测 native Result 上的可选字段。
+// 探测出 cancel/abort/side_effects 任何一个就使用它们；探测失败或全部为零值视为
+// legacy reading（SideEffects 空 = possible 兜底）。
+func probeNativeResultExtras(res *exec.Result) (nativeResultExtras, error) {
+	if res == nil {
+		return nativeResultExtras{}, fmt.Errorf("nil result")
+	}
+	v := reflect.ValueOf(res).Elem()
+	if !v.IsValid() {
+		return nativeResultExtras{}, fmt.Errorf("invalid result")
+	}
+	// 用 JSON 做结构化探测：这一路兼容 deps 升级（增字段）、不增字段两种情形。
+	data, err := json.Marshal(v.Interface())
+	if err != nil {
+		return nativeResultExtras{}, err
+	}
+	var out nativeResultExtras
+	if err := json.Unmarshal(data, &out); err != nil {
+		return nativeResultExtras{}, err
+	}
+	return out, nil
 }

 type eventTap struct {
@@ -47,3 +47,85 @@ func TestBuildOptionsBashPermissionGating(t *testing.T) {
 		t.Fatal("missing model must fail-closed")
 	}
 }
+
+func TestConfigIdentityReflectsToolAndClient(t *testing.T) {
+	base := Config{
+		Client:       &llm.Client{BaseURL: "http://x", APIKey: "k"},
+		DefaultModel: "m1", MaxTurns: 5, Timeout: time.Minute,
+		Exec: &tools.ExecConfig{Bash: "/bin/bash"},
+	}
+	otherExec := base
+	otherExec.Exec = &tools.ExecConfig{Bash: "/bin/zsh"}
+	if base.Identity() == otherExec.Identity() {
+		t.Fatal("Exec tool-config change must change Config.Identity")
+	}
+	otherClient := base
+	otherClient.Client = &llm.Client{BaseURL: "http://y", APIKey: "k"}
+	if base.Identity() == otherClient.Identity() {
+		t.Fatal("client change must change Config.Identity")
+	}
+	same := base
+	if base.Identity() != same.Identity() {
+		t.Fatal("identical config must keep Identity stable")
+	}
+}
+
+func TestRegisterFullAllowDenyTakesEffectOnNextResolve(t *testing.T) {
+	r := squad.NewBackendResolver()
+	cfg := Config{
+		Client:       &llm.Client{BaseURL: "http://x", APIKey: "k"},
+		DefaultModel: "m1", MaxTurns: 1, Timeout: time.Second,
+		Exec: &tools.ExecConfig{Bash: "/bin/bash"},
+	}
+	RegisterFull(r, cfg)
+
+	allowRole := squad.RoleSpec{Role: "eng", Backend: "agent-native", Model: "m1", BashPermission: squad.BashPermissionAllow}
+	denyRole := squad.RoleSpec{Role: "eng", Backend: "agent-native", Model: "m1", BashPermission: squad.BashPermissionDeny}
+
+	eAllow, _, err := r.Resolve(allowRole, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	eDeny, _, err := r.Resolve(denyRole, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if eAllow == eDeny {
+		t.Fatal("allow→deny must not reuse cached executor on next Resolve")
+	}
+	// Identity of role configs must diverge so cache keys differ.
+	if squad.ConfigIdentity(allowRole) == squad.ConfigIdentity(denyRole) {
+		t.Fatal("role ConfigIdentity must include BashPermission")
+	}
+}
+
+func TestRegisterFullToolConfigReplacementTakesEffect(t *testing.T) {
+	r := squad.NewBackendResolver()
+	cfg1 := Config{
+		Client:       &llm.Client{BaseURL: "http://x", APIKey: "k"},
+		DefaultModel: "m1", MaxTurns: 1, Timeout: time.Second,
+		Exec: &tools.ExecConfig{Bash: "/bin/bash"},
+	}
+	cfg2 := Config{
+		Client:       &llm.Client{BaseURL: "http://x", APIKey: "k"},
+		DefaultModel: "m1", MaxTurns: 1, Timeout: time.Second,
+		Exec: &tools.ExecConfig{Bash: "/bin/zsh"},
+	}
+	if cfg1.Identity() == cfg2.Identity() {
+		t.Fatal("precondition: tool configs must differ in Identity")
+	}
+	RegisterFull(r, cfg1)
+	role := squad.RoleSpec{Role: "eng", Backend: "agent-native", Model: "m1", BashPermission: squad.BashPermissionAllow}
+	e1, _, err := r.Resolve(role, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	RegisterFull(r, cfg2)
+	e2, _, err := r.Resolve(role, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if e1 == e2 {
+		t.Fatal("RegisterFull tool-config replacement must invalidate cache")
+	}
+}
@@ -4,6 +4,7 @@ import (
 	"context"
 	"os"
 	"path/filepath"
+	"strings"
 	"testing"
 	"time"

@@ -127,7 +128,7 @@ func TestEngineRunsOnSQLiteStore(t *testing.T) {
 		}
 	}
 	backends := squad.NewBackendResolver()
-	backends.RegisterInstance("script", stubExecutor{result: &squad.ExecResult{Success: false, Logs: "ok", Patches: map[string][]byte{"output.md": []byte(goodReportLite), "notes.md": []byte(goodReportLite), "fields.json": []byte("{}")}}})
+	backends.RegisterInstance("script", stubExecutor{result: &squad.ExecResult{Success: true, Logs: "ok", Patches: map[string][]byte{"output.md": []byte(goodReportLite), "notes.md": []byte(goodReportLite), "fields.json": []byte("{}")}}})
 	blueprints := squad.NewBlueprintRegistry()
 	if err := blueprints.Register(squad.Blueprint{
 		Key: "demo", Title: "入库演示", Match: []string{"入库演示"},
@@ -185,4 +186,78 @@ func TestEngineRunsOnSQLiteStore(t *testing.T) {
 	}
 }

+// TestEngineFailedResultWithCompletePatchesStaysFailedOnSQLiteStore：F05/U3B 的
+// 不变式在 SQLite-store 集成层同样成立——executor 报失败就是失败，patches
+// （交付物）齐全也不得把运行翻成 completed：文件只是诊断信息，不是权威结果。
+func TestEngineFailedResultWithCompletePatchesStaysFailedOnSQLiteStore(t *testing.T) {
+	runStore, err := OpenSQLiteStore(filepath.Join(t.TempDir(), "runs.db"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer runStore.Close()
+	dir := t.TempDir()
+	reg := squad.NewRoleRegistry()
+	if err := reg.Register(squad.RoleSpec{Role: "researcher", Backend: "script"}); err != nil {
+		t.Fatal(err)
+	}
+	backends := squad.NewBackendResolver()
+	// 失败结果但携带全套交付物：旧语义会把它提升为成功，U3B 起必须保持失败。
+	backends.RegisterInstance("script", stubExecutor{result: &squad.ExecResult{
+		Success: false, Logs: "stub-refused",
+		Patches: map[string][]byte{
+			"output.md":   []byte(goodReportLite),
+			"notes.md":    []byte(goodReportLite),
+			"fields.json": []byte("{}"),
+		},
+	}})
+	blueprints := squad.NewBlueprintRegistry()
+	if err := blueprints.Register(squad.Blueprint{
+		Key: "faildemo", Title: "失败演示", Match: []string{"失败演示"},
+		Steps: []squad.StepSpec{{ID: "s1", Role: "researcher", Title: "调研", Targets: []string{"output.md"}}},
+	}); err != nil {
+		t.Fatal(err)
+	}
+	eng, err := squad.NewEngine(squad.Options{
+		DataDir: dir, Registry: reg, Blueprints: blueprints,
+		Planner:  squad.NewRulePlanner(),
+		Backends: backends, Sink: squad.NopSink{},
+		Store: runStore,
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	res, err := eng.Run(context.Background(), squad.Goal{Text: "做一个失败演示", HitlMode: squad.HitlAuto}, squad.RunOptions{})
+	if err != nil {
+		t.Fatalf("run: %v", err)
+	}
+	if res.Status == squad.RunCompleted {
+		t.Fatalf("failed executor result must not complete (patches are diagnostic only)")
+	}
+	if res.Status != squad.RunFailed {
+		t.Fatalf("status = %s, want failed", res.Status)
+	}
+	if res.Plan == nil || res.Plan.Steps[0].Status != "failed" {
+		t.Fatalf("s1 should be failed, plan = %+v", res.Plan)
+	}
+	if !strings.Contains(res.Error, "stub-refused") {
+		t.Fatalf("failure reason not preserved: %q", res.Error)
+	}
+	// 同一断言穿过 Store 端口：SQLite 里落盘的也是失败，而非 completed。
+	st, err := runStore.LoadRun(res.WorkDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if st.Plan.Workflow.Status != "failed" {
+		t.Fatalf("persisted workflow status = %s, want failed", st.Plan.Workflow.Status)
+	}
+	if got := st.Plan.Workflow.Steps[0]; got.Status != "failed" || !strings.Contains(got.Error, "stub-refused") {
+		t.Fatalf("persisted step = %+v", got)
+	}
+	// patches 确已物化（文件齐全），依然不翻转状态：权威是执行结果，不是文件。
+	sub := filepath.FromSlash(st.Plan.Workflow.Steps[0].SubWorkspace)
+	if _, err := os.Stat(filepath.Join(res.WorkDir, sub, "output.md")); err != nil {
+		t.Fatalf("diagnostic patch should still be materialized: %v", err)
+	}
+}
+
 const goodReportLite = "# 测试交付物\n\n## 概述\n\n这是由脚本后端生成的合格交付物，长度超过五十个字符以满足最小质检要求。\n"
@@ -1,24 +1,26 @@
 # 通用规范采纳记录

- 项目/负责人：Agent-Squad / 仓库所有者 weichen
+- 项目/负责人：Agent-Squad / weichen
 - 状态：active
 - 规范位置：`deps/development-guidelines`
 - 规范性文档目录：`policy/`
- 过程文档位置：`docs/records/`（项目内固定目录豁免。个人维护的编排库，任务主记录仍在 Gitea Issue/PR。过程记录不进规范库，也不写成与 Issue 平行的第二状态。）
- 版本权威：父仓 gitlink；可读标签：无（内容版本 `0.1.0-draft.6`，提交 `26b562d2e34fbc560711296a309c91c74ffc55b1`）
- 批准/生效依据：本仓接入提交
- 替代范围和迁移说明：本仓此前没有根 `AGENTS.md`。Agent-Native 子模块已从仓库根迁到 `deps/agent-native`，迁移说明见 `docs/records/2026-09-29-guidelines-and-observability.md`。
+- 过程文档位置：任务卡在 `.task/`（`.task/feature-governance-adoption.md`）；任务主记录在 Gitea PR 中维护。既有过程记录仍按项目内固定目录豁免留在 `docs/records/`。过程记录不进规范库，也不写成与 Issue/PR 平行的第二状态。
+- 版本权威：本仓 gitlink deps/development-guidelines @ 2688fd2be6a5（完整 `2688fd2be6a52c0bf110e7e6366b6f291ab5e4eb`，内容版本 `0.1.0-draft.12`，无已发布标签）
+- 批准/生效依据：`feature/governance-adoption PR（Agent-Squad）`
+- 替代范围和迁移说明：本次替代 2026-09-29 采纳的固定版本 `26b562d2e34f`（内容版本 `0.1.0-draft.6`），规则条目按 `0.1.0-draft.12` 重新核对。新增 `AGENT-EXECUTION.md` 执行路由、`deps/development-manager` gitlink（`bd3a7de57a3a`）、`scripts/dm`、`scripts/agent-exec`、`scripts/task`、`scripts/hooks/pre-commit`、`.gitea/workflows/governance.yml` 与任务卡模板。不改业务代码，不改既有 `.gitea/workflows/ci.yml`。此前迁移说明见 `docs/records/2026-09-29-guidelines-and-observability.md`。

 | 规范/工作流 | 适用条件与范围 | 采纳或不适用理由 | 执行方式/证据 |
 |---|---|---|---|
 | core/baseline.md | 日常工作 | 编排库对外有失败关闭语义 | manual |
 | workflows/delivery.md、debugging.md、handoff.md | 交付、排障与交接 | 跨仓组合需要可交接记录 | manual |
 | standards/collaboration.md | 并行共同起点与组合升级 | 父仓 gitlink 锁定 Agent-Native 与基础库 | manual |
-| standards/verification.md | 验证与结论 | `GOWORK=off go test` | manual；命令见 PROJECT.md |
+| standards/merge-queue.md | 多个就绪任务的最终同步、CI、Review 与合入调度 | 多个 agent 单元并行开发，合入须串行同步；协调者为仓库所有者 | manual |
+| standards/verification.md | 验证与结论 | `GOWORK=off go test` | manual；命令见 PROJECT.md（含 `scripts/agent-exec validate --repo .`、`scripts/dm governance check --repo .`） |
 | workflows/design-verification.md、standards/testability.md、standards/failure-containment.md | 规划、执行、HITL | conformance 覆盖公开 API | manual |
 | standards/module-contracts.md | squad 与 agent-native 边界 | 依赖方向只允许 squad → agent-native | manual |
 | standards/observability.md | HTTP 信封、trace、事件 | 接入错误码库与日志库 | manual |
-| standards/agent-execution.md | 角色绑定的执行后端 | 本库不实现 Agent | manual |
+| standards/agent-execution.md | 工程 Agent CLI 执行路由、会话隔离与受控集成 | 凡允许 AI 协助工程变更必须采纳；路由配置见 AGENT-EXECUTION.md；本库不实现 Agent | 执行方式：AGENT-EXECUTION.md + scripts/agent-exec（validate / admit）；证据：`governance check + 正负例（.task/feature-governance-adoption.md）` |
+| standards/tool-adaptation.md | 为具体工具创建规则与钩子 | 本次引入 `scripts/hooks/pre-commit`（启用由 host/人执行，`core.hooksPath`） | manual；启用后以本仓 pre-commit 检查为证据 |
 | standards/multi-repo.md | 多级 submodule | 本仓 replace 是编译时的唯一依赖解析 | manual |
 | standards/data-reliability.md | 运行目录、HITL 恢复、幂等回应 | 磁盘为权威 | manual |
 | standards/security.md | bash 权限挂起、脱敏 | 事件落盘沿用现有脱敏；凭证走环境变量 | manual |
@@ -31,14 +33,17 @@

 | 原规则/章节 | 项目采用方式 | 原因与风险 | 负责人/复查条件 |
 |---|---|---|---|
-| documentation.md 过程仓默认独立仓库 | 过程记录放 `docs/records/` | 同 Agent-Native：避免个人项目再维护一个过程仓。风险是证据与代码同库 | 多人协作或对外发布时复查 |
+| documentation.md 过程仓默认独立仓库 | 其他过程记录放 `docs/records/`（既有过程记录的项目内固定目录豁免） | 同 Agent-Native：避免个人项目再维护一个过程仓。风险是过程记录与代码同库 | 多人协作或对外发布时复查 |
+| documentation.md 过程仓默认独立仓库 | 任务卡与交付证据放 `.task/`（模板 `.task/templates/task-card.md`） | 任务状态仍只认 Gitea Issue/PR，`.task/` 不写成平行状态；风险是任务卡与代码同库 | 多人协作或对外发布时复查 |
 | multi-repo.md 组件说明不手写第二套 SHA | PROJECT.md 只写路径与角色，SHA 以 `git submodule status` 为准 | 手抄 SHA 会与 gitlink 漂移 | 每次依赖升级核对脚本 |

 ## 本次升级或首次接入验证

- `deps/development-guidelines` 固定 `26b562d2e34fbc560711296a309c91c74ffc55b1`。
- 执行方式为 manual。
- 回退该规范挂载时，同时确认 Agent-Native 子模块路径是否已迁移；只还原规范 gitlink 不会把 `deps/agent-native` 移回根目录。
+- `deps/development-guidelines` 固定 `2688fd2be6a52c0bf110e7e6366b6f291ab5e4eb`（内容版本 `0.1.0-draft.12`）；`deps/development-manager` 固定 `bd3a7de57a3a9e1a4495b94c5fa5c08d71e7be8c`。
+- 工程 Agent 执行：`AGENT-EXECUTION.md` 为唯一路由权威；AI 实施单元只经 `scripts/agent-exec run` 执行，并经 `scripts/agent-exec apply` 集成。`scripts/agent-exec` 与 `scripts/task` 只是转调 `scripts/dm` 的薄入口。
+- 登记为 active 的前置证据：一次成功执行正例（有效 `completed` 收据经 transport/admission 集成）；缺配置、无效终态收据、越权路径、未知副作用四类负例均被拒绝或 fail-stop。证据按 `.task/feature-governance-adoption.md` 记录。
+- 入口核验：`scripts/dm governance check --repo .` 与 `scripts/agent-exec validate --repo .`（PROJECT.md「验证」节）。CI 由 `.gitea/workflows/governance.yml` 执行；既有 `ci.yml` 未改。
+- 回退该规范挂载时，恢复上一已批准 gitlink `26b562d2e34f`，并同时还原本次新增的治理脚本与 `governance.yml`；如已启用 `core.hooksPath`，需一并关闭。只还原规范 gitlink 不会把 `deps/agent-native` 移回根目录。

 ### 2026-09-29 追加：agent-intent 可选意图闸门

@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+# 共享 Agent 执行器薄入口：转调 scripts/dm 的 agent-exec 子命令（validate / run / apply / admit / cleanup）。
+set -euo pipefail
+exec "$(dirname "$0")/dm" agent-exec "$@"
@@ -0,0 +1,55 @@
+#!/usr/bin/env bash
+# 共享 dm-server 薄入口：读取 deps/development-manager 的 gitlink 钉住提交，
+# 按该提交构建并缓存 dm-server，再 exec 其子命令（agent-exec、task、governance）。执行语义与退出码以该二进制为准。
+set -euo pipefail
+
+root=$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd -P)
+dm="$root/deps/development-manager"
+
+die() { echo "dm: $*" >&2; exit 2; }
+
+# git 钩子会导出仓库本地变量（GIT_DIR、GIT_INDEX_FILE 等），它们会让 `git -C "$dm"` 解析到外层仓库。
+# 所有针对 DM 检出的 git 与 go 调用都在清除这些变量后执行；读取外层 gitlink 与最终 exec 仍保留继承环境。
+dm_local_env="GIT_DIR GIT_WORK_TREE GIT_INDEX_FILE GIT_PREFIX GIT_OBJECT_DIRECTORY GIT_ALTERNATE_OBJECT_DIRECTORIES GIT_COMMON_DIR GIT_GRAFT_FILE GIT_NAMESPACE GIT_CONFIG GIT_CONFIG_PARAMETERS GIT_CONFIG_COUNT GIT_SHALLOW_FILE GIT_IMPLICIT_WORK_TREE $(git rev-parse --local-env-vars 2>/dev/null || true)"
+dm_clean_env() { for v in $dm_local_env; do unset "$v"; done; }
+dmgit() ( dm_clean_env; git -C "$dm" "$@" )
+
+# 1. 钉住版本只来自索引中的 gitlink（mode 160000），不读取工作区文件。
+entry=$(git -C "$root" ls-files -s -- deps/development-manager 2>/dev/null || true)
+[ -n "$entry" ] || die "索引中缺少 deps/development-manager 的 gitlink，无法确定钉住的 DM 版本"
+read -r mode pin _stage _path <<<"$entry"
+[ "$mode" = 160000 ] || die "deps/development-manager 不是 gitlink（mode=$mode）"
+[[ "$pin" =~ ^[0-9a-f]{40}$ ]] || die "gitlink 提交格式无效：$pin"
+
+# 2. 检出必须存在且确为 DM 独立仓库（避免 git -C 误解析到外层仓库），HEAD 必须等于钉住提交。
+[ -e "$dm/.git" ] || die "deps/development-manager 未检出；请按钉住提交 $pin 检出该 gitlink"
+top=$(dmgit rev-parse --show-toplevel 2>/dev/null || true)
+[ -n "$top" ] && [ "$top" = "$(cd "$dm" && pwd -P)" ] \
+  || die "deps/development-manager 不是独立检出；请按钉住提交 $pin 检出该 gitlink"
+head=$(dmgit rev-parse HEAD 2>/dev/null || true)
+[ "$head" = "$pin" ] \
+  || die "deps/development-manager HEAD（${head:-未知}）与钉住提交 $pin 不一致；请按钉住提交检出"
+
+# 3. 检出工作树必须干净（忽略嵌套 submodule 状态）。
+dirty=$(dmgit status --porcelain --ignore-submodules=all) \
+  || die "无法读取 deps/development-manager 的工作树状态"
+[ -z "$dirty" ] || die "deps/development-manager 存在未提交改动，拒绝运行（请先清理该检出）"
+
+# 4. 二进制按钉住提交缓存；缺失时在缓存目录内临时构建，完成后原子改名，避免留下半成品。
+cache="${DM_CACHE:-${XDG_CACHE_HOME:-$HOME/.cache}/development-manager/dm-server}"
+bin="$cache/dm-server-$pin"
+if [ ! -x "$bin" ]; then
+  command -v go >/dev/null 2>&1 || die "缺少 go，无法构建 dm-server（$pin）"
+  mkdir -p "$cache"
+  cache=$(cd "$cache" && pwd -P)
+  bin="$cache/dm-server-$pin"
+  work=$(mktemp -d "$cache/build-$pin.XXXXXX")
+  if ! (dm_clean_env; cd "$dm" && GOWORK=off go build -trimpath -o "$work/dm-server" ./cmd/dm-server); then
+    rm -rf "$work"
+    die "构建 dm-server 失败（$pin）"
+  fi
+  mv -f "$work/dm-server" "$bin"
+  rm -rf "$work"
+fi
+
+exec "$bin" "$@"
@@ -0,0 +1,16 @@
+#!/usr/bin/env bash
+# 本地检查（pre-commit）：集成分支直提拒绝 + 分支命名 + 本地完整性核对。权威准入在 PR（baseline.md:26）。
+branch=$(git rev-parse --abbrev-ref HEAD)
+case "$branch" in
+  main|master)
+    echo "pre-commit: 拒绝——集成分支 ${branch} 不得直接提交，走 feature 分支 + PR（collaboration.md）" >&2
+    exit 1 ;;
+esac
+echo "$branch" | grep -qE '^(feature|fix|hotfix|release)/[A-Za-z0-9._-]+$' || {
+  echo "pre-commit: 拒绝——分支名须为 feature|fix|hotfix|release/<slug>，当前：${branch}" >&2
+  exit 1
+}
+root=$(git rev-parse --show-toplevel)
+# admit 只是本地完整性/篡改证据（SELF）：同一 OS 用户可读取本地 HMAC 密钥并绕过本钩子，
+# 它不是 GATE，也不是可信身份。缺失或无效的本地记录仍拒绝；交付信任来自精确 HEAD 的 CI 或独立 REPLAY。
+"$root/scripts/agent-exec" admit --repo "$root"
@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+# 通用任务入口：转调 scripts/dm 的 task 子命令（triage / start / pr-check）。
+set -euo pipefail
+exec "$(dirname "$0")/dm" task "$@"
@@ -2,12 +2,18 @@ package squad

 import (
 	"context"
+	"encoding/json"
 	"os"
 	"path/filepath"
 	"sync"
 	"testing"
+	"time"
 )

+// jsonForTest 包装 json.Marshal 让测试里构型参数不拉锁。是仅为占用 import 走
+// 公共路径，不污染生产代码。
+func jsonForTest(v any) ([]byte, error) { return json.Marshal(v) }
+
 // perStepRecordingBackend 按子工作区归因计数（steps 的 role→tasks/NN_role 布局）。
 type perStepRecordingBackend struct {
 	mu    sync.Mutex
@@ -59,6 +65,9 @@ func newAttemptTestEngine(t *testing.T, backend *perStepRecordingBackend, dir st
 	return eng
 }

+// flipToInterrupted 把该步骤手动置为 running，同时清除 trusted terminal
+// record（模拟「崩溃于执行中」状态）。清除让该路径能够测试 F05 走的
+// 「无 record」分支。
 func flipToInterrupted(t *testing.T, workDir string, stepIdx int) {
 	t.Helper()
 	st, err := loadRunState(workDir)
@@ -75,6 +84,16 @@ func flipToInterrupted(t *testing.T, workDir string, stepIdx int) {
 	if err := st.save(); err != nil {
 		t.Fatal(err)
 	}
+	// 同时清除该步骤的 trusted terminal record（模拟未达记录写入点崩溃）。
+	terminals := loadStepTerminalsForWorkDir(workDir)
+	if terminals != nil {
+		stepID := st.Plan.Workflow.Steps[stepIdx].ID
+		if _, ok := terminals[stepID]; ok {
+			delete(terminals, stepID)
+			data, _ := jsonForTest(terminals)
+			_ = os.WriteFile(filepath.Join(workDir, "step_terminals.json"), data, 0o644)
+		}
+	}
 }

 // TestAttemptIDFormat：attempt id 稳定落盘且形如 runID/stepID/N（Q1 WP3）。
@@ -98,8 +117,15 @@ func TestAttemptIDFormat(t *testing.T) {
 }

 // TestKilledAtSuperstepBoundarySideEffectsOnce：杀进程于 superstep 边界后恢复，
-// 已完成步骤不重新执行（执行次数=1）；running 且产物齐全的步骤按 attempt 判定
-// 完成、不 rerun——Q1 验收第一条（WP3）。
+// 已完成步骤不重新执行（执行次数=1）。
+//
+// F05/U3b：running 崩溃、文件齐全、却无 trusted terminal record（U3b 前被视为
+// “副作用真发生”自动完成；U3b 后仅记录 step.Status=needs_reconciliation，
+// 后台不再自动 rerun/完成。本测试验证该迁变后的语义不变量：
+//
+//   - running 且无 record：需要 reconciliation，不隐默 rerun；
+//   - file presence 不替 trusted record 完成步骤。
+//   - 若事先写入 trusted record（手工合成），即便文件缺失也恢复为 completed。
 func TestKilledAtSuperstepBoundarySideEffectsOnce(t *testing.T) {
 	backend := &perStepRecordingBackend{}
 	dir := t.TempDir()
@@ -125,22 +151,78 @@ func TestKilledAtSuperstepBoundarySideEffectsOnce(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	if res.Status != RunCompleted {
-		t.Fatalf("resume: %+v", res)
+	// F05：需 reconciliation（auto 模式），不等于 completed。
+	if res.Status != RunWaitingReconciliation {
+		t.Fatalf("want waiting_reconciliation (auto mode), got %s", res.Status)
 	}
+	if res.Plan == nil || len(res.Plan.Steps) == 0 {
+		t.Fatal("missing plan")
+	}
+	var recon int
 	for _, s := range res.Plan.Steps {
-		if s.Status != "completed" && s.Status != "skipped" {
-			t.Fatalf("step %s not finished: %s", s.ID, s.Status)
+		if s.ID == "s2" && s.Status == StepStatusNeedsReconciliation {
+			recon++
 		}
+	}
+	if recon != 1 {
+		t.Fatalf("step s2 not in needs_reconciliation (got plan=%+v)", res.Plan.Steps)
+	}
+	// 重要：重复执行的次数不应发生隐默 rerun（本测试用 auto 模式运行走到 reconciliation）。
+	for _, s := range res.Plan.Steps {
 		got := backend.countOf(filepath.Base(filepath.FromSlash(s.SubWorkspace)))
 		if got != before[s.ID] {
-			t.Fatalf("step %s executed %d times after kill (before=%d)——已完成/running 步骤被重复执行", s.ID, got, before[s.ID])
+			t.Fatalf("step %s executed %d times after kill (before=%d)——running 步骤被隐默 rerun", s.ID, got, before[s.ID])
 		}
 	}
 }

-// TestInterruptedMissingProductsRetries：running 崩溃且产物缺失时按 auto 一次重试（WP3）。
-func TestInterruptedMissingProductsRetries(t *testing.T) {
+// TestTrustedRecordCompletesMissingFiles：trusted terminal record 存在 + 文件缺失
+// → 恢复为 completed（files 仍作诊断，state 是权威）。
+func TestTrustedRecordCompletesMissingFiles(t *testing.T) {
+	backend := &perStepRecordingBackend{}
+	dir := t.TempDir()
+	eng := newAttemptTestEngine(t, backend, dir)
+	res, err := eng.Run(context.Background(), Goal{Text: "trusted record 完成", HitlMode: HitlAuto}, RunOptions{})
+	if err != nil || res.Status != RunCompleted {
+		t.Fatalf("setup run: %v %+v", err, res)
+	}
+	// 制造「第二步走 running 但 trusted record 已存在」的场景：先写入 record，再
+	// flip 到 running + 删 file。
+	st, err := loadRunState(res.WorkDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+	sub2 := filepath.FromSlash(st.Plan.Workflow.Steps[1].SubWorkspace)
+	flipToInterrupted(t, res.WorkDir, 1)
+	if err := os.Remove(filepath.Join(res.WorkDir, sub2, "output.md")); err != nil {
+		t.Fatal(err)
+	}
+	// 手工写入 trusted record（这能覆盖「后端状态已落盘但文件被外部删」该返回
+	// 已完成的合法语义）。
+	terminals := map[string]stepTerminalRecord{
+		"s2": {StepID: "s2", AttemptID: st.Plan.Workflow.Steps[1].AttemptID, Terminal: "completed", SideEffects: string(ExecSideEffectsNone), At: time.Now()},
+	}
+	data, _ := jsonForTest(terminals)
+	if err := os.WriteFile(filepath.Join(res.WorkDir, "step_terminals.json"), data, 0o644); err != nil {
+		t.Fatal(err)
+	}
+	before := backend.countOf(filepath.Base(sub2))
+	eng2 := newAttemptTestEngine(t, backend, dir)
+	res, err = eng2.Resume(context.Background(), res.RunID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res.Status != RunCompleted {
+		t.Fatalf("want completed via trusted record, got %s", res.Status)
+	}
+	if got := backend.countOf(filepath.Base(sub2)); got != before {
+		t.Fatalf("trusted-record step reran (before=%d after=%d)——record 需被信任为 completed", before, got)
+	}
+}
+
+// TestInterruptedMissingProductsEntersReconciliation：F05/U3b：running 崩溃 + 无
+// trusted record + 文件缺失 → auto 模式不再隐默 rerun，进入 waiting_reconciliation。
+func TestInterruptedMissingProductsEntersReconciliation(t *testing.T) {
 	backend := &perStepRecordingBackend{}
 	dir := t.TempDir()
 	eng := newAttemptTestEngine(t, backend, dir)
@@ -148,7 +230,6 @@ func TestInterruptedMissingProductsRetries(t *testing.T) {
 	if err != nil || res.Status != RunCompleted {
 		t.Fatalf("setup run: %v %+v", err, res)
 	}
-	// 第二步停 running 并删掉其交付物（副作用未发生的形态）。
 	flipToInterrupted(t, res.WorkDir, 1)
 	st, err := loadRunState(res.WorkDir)
 	if err != nil {
@@ -158,16 +239,24 @@ func TestInterruptedMissingProductsRetries(t *testing.T) {
 	if err := os.Remove(filepath.Join(res.WorkDir, sub, "output.md")); err != nil {
 		t.Fatal(err)
 	}
+	before := backend.countOf(filepath.Base(sub))
 	eng2 := newAttemptTestEngine(t, backend, dir)
 	res, err = eng2.Resume(context.Background(), res.RunID)
 	if err != nil {
 		t.Fatal(err)
 	}
-	if res.Status != RunCompleted {
-		t.Fatalf("resume: %+v", res)
+	if res.Status != RunWaitingReconciliation {
+		t.Fatalf("want waiting_reconciliation, got %s", res.Status)
 	}
-	got := backend.countOf(filepath.Base(sub))
-	if got != 2 {
-		t.Fatalf("missing-product step executed %d times, want 2（auto 一次重试）", got)
+	// 后续 Resume 仍旧是 reconciliation，不再 rerun。
+	res2, err := eng2.Resume(context.Background(), res.RunID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res2.Status != RunWaitingReconciliation {
+		t.Fatalf("second resume shouldn't rerun, got %s", res2.Status)
+	}
+	if got := backend.countOf(filepath.Base(sub)); got != before {
+		t.Fatalf("missing-files step reran (before=%d after=%d)——record 缺失不应 rerun", before, got)
 	}
 }
@@ -1,7 +1,9 @@
 package squad

 import (
+	"context"
 	"fmt"
+	"strings"
 	"sync"

 	"error-code-system/errcode"
@@ -16,11 +18,20 @@ type BackendFactory func(role RoleSpec) (Executor, error)
 // agent-native 自研内核，产品也可注册 "pi-agent" 等外部智能体工厂。
 // 未知名字 fail-closed，绝不静默落 stub。角色声明 MaxConcurrentTasks 时
 // 以并发限流包装实现对象级串行（每个角色一份缓存实例，限流跨步骤共享）。
+//
+// 缓存键含 ConfigIdentity 与工厂世代，故 BashPermission/工具配置收紧或工厂
+// 替换在下一次 Resolve 即生效。并发 gate 按「后端名+角色」独立于执行器版本
+// 持有：配置换代不会另起限流器而绕过角色上限。
+//
+// 边界：已在飞的任务不会被回滚或取消；收紧权限/降低并发只影响后续 Resolve
+// 与后续 Acquire。需要即时撤权的宿主应在每次工具调用前再校验。
 type BackendResolver struct {
 	mu         sync.RWMutex
 	factories  map[string]BackendFactory
 	underlying map[string]Executor // 无工厂直挂的后端实例
-	wrapped    map[string]Executor // 角色（含模型绑定）→ 已包装实例缓存
+	wrapped    map[string]Executor // 执行器配置身份 → 已包装实例缓存
+	factoryGen map[string]uint64   // Register/RegisterInstance 世代，替换即递增
+	gates      map[string]*roleConcurrencyGate
 }

 // NewBackendResolver 创建解析器并注册内置后端。
@@ -29,23 +40,40 @@ func NewBackendResolver() *BackendResolver {
 		factories:  make(map[string]BackendFactory),
 		underlying: make(map[string]Executor),
 		wrapped:    make(map[string]Executor),
+		factoryGen: make(map[string]uint64),
+		gates:      make(map[string]*roleConcurrencyGate),
 	}
 	r.RegisterInstance("stub", &StubExecutor{})
 	return r
 }

 // Register 注册后端工厂（名字如 "agent-native"、"pi-agent"）。
+// 替换同名工厂会使该后端下已缓存的包装执行器失效。
 func (r *BackendResolver) Register(name string, f BackendFactory) {
 	r.mu.Lock()
 	defer r.mu.Unlock()
 	r.factories[name] = f
+	r.invalidateLocked(name)
 }

 // RegisterInstance 直接注册后端实例（忽略角色级绑定）。
+// 替换同名实例会使该后端下已缓存的包装执行器失效；"stub" 语义保持不变。
 func (r *BackendResolver) RegisterInstance(name string, b Executor) {
 	r.mu.Lock()
 	defer r.mu.Unlock()
 	r.underlying[name] = b
+	r.invalidateLocked(name)
+}
+
+func (r *BackendResolver) invalidateLocked(name string) {
+	r.factoryGen[name]++
+	prefix := name + "\x00"
+	for k := range r.wrapped {
+		if strings.HasPrefix(k, prefix) {
+			delete(r.wrapped, k)
+		}
+	}
+	// gates 按角色身份保留：换工厂/配置不得丢掉进行中的并发计数。
 }

 // ListBackends 返回可用后端名（排序稳定）。
@@ -73,8 +101,8 @@ func (r *BackendResolver) ListBackends() []string {
 }

 // Resolve 解析角色的执行后端：Role.Backend → Options.DefaultBackend → "stub"。
-// MaxConcurrentTasks > 1 时以 agent-native 并发限流包装（per-role 缓存，跨步骤共享；
-// 创建段持写锁防并行首解析竞态——否则并发步骤会各拿一份限流器）。
+// MaxConcurrentTasks >= 1 时以共享角色 gate 包装（跨执行器版本共用；
+// 创建段持写锁防并行首解析竞态——否则并发步骤会各拿一份执行器）。
 func (r *BackendResolver) Resolve(role RoleSpec, defaultBackend string) (Executor, string, error) {
 	name := role.Backend
 	if name == "" {
@@ -83,30 +111,44 @@ func (r *BackendResolver) Resolve(role RoleSpec, defaultBackend string) (Executo
 	if name == "" {
 		name = "stub"
 	}
-	cacheKey := name + "\x00" + role.Role + "\x00" + role.Model

 	// 后端名快照在锁外获取（Resolve 持锁期间再调 ListBackends 会自死锁）。
 	names := r.ListBackends()
+
 	r.mu.RLock()
+	gen := r.factoryGen[name]
+	cacheKey := resolveCacheKey(name, role, gen)
 	cached, ok := r.wrapped[cacheKey]
-	factory, hasFactory := r.factories[name]
-	under, hasUnder := r.underlying[name]
 	r.mu.RUnlock()
 	if ok {
+		if max := role.MaxConcurrentTasks; max >= 1 {
+			r.mu.Lock()
+			if g := r.gates[limiterKey(name, role.Role)]; g != nil {
+				g.setMax(max)
+			}
+			r.mu.Unlock()
+		}
 		return cached, name, nil
 	}

 	r.mu.Lock()
 	defer r.mu.Unlock()
+	gen = r.factoryGen[name]
+	cacheKey = resolveCacheKey(name, role, gen)
 	// 双检：等写锁期间可能已被并发解析填充。
 	if b, ok := r.wrapped[cacheKey]; ok {
+		if max := role.MaxConcurrentTasks; max >= 1 {
+			if g := r.gates[limiterKey(name, role.Role)]; g != nil {
+				g.setMax(max)
+			}
+		}
 		return b, name, nil
 	}

 	var b Executor
-	if hasUnder {
+	if under, hasUnder := r.underlying[name]; hasUnder {
 		b = under
-	} else if hasFactory {
+	} else if factory, hasFactory := r.factories[name]; hasFactory {
 		var err error
 		b, err = factory(role)
 		if err != nil {
@@ -117,9 +159,114 @@ func (r *BackendResolver) Resolve(role RoleSpec, defaultBackend string) (Executo
 	}

 	// MaxConcurrentTasks>=1 即包装限流（1 = 对象级串行；0 = 不限）。
+	// gate 按角色身份共享，与 ConfigIdentity / 工厂世代无关。
 	if max := role.MaxConcurrentTasks; max >= 1 {
-		b = WrapWithConcurrencyLimit(b, max)
+		g := r.gateLocked(name, role.Role, max)
+		b = &gatedExecutor{inner: b, gate: g}
 	}
 	r.wrapped[cacheKey] = b
 	return b, name, nil
 }
+
+func resolveCacheKey(name string, role RoleSpec, gen uint64) string {
+	return name + "\x00" + role.Role + "\x00" + ConfigIdentity(role) + "\x00" +
+		fmt.Sprintf("%d", gen) + "\x00" + limiterStamp(role.MaxConcurrentTasks)
+}
+
+func limiterKey(backend, role string) string {
+	return backend + "\x00" + role
+}
+
+func (r *BackendResolver) gateLocked(backend, role string, max int) *roleConcurrencyGate {
+	key := limiterKey(backend, role)
+	g, ok := r.gates[key]
+	if !ok {
+		g = newRoleConcurrencyGate(max)
+		r.gates[key] = g
+		return g
+	}
+	g.setMax(max)
+	return g
+}
+
+// roleConcurrencyGate 是按角色共享的并发上限。
+// 执行器配置换代仍复用同一 gate，故旧版本在飞任务计入同一上限。
+type roleConcurrencyGate struct {
+	mu     sync.Mutex
+	max    int
+	active int
+	notify chan struct{} // 关闭并替换以唤醒等待者
+}
+
+func newRoleConcurrencyGate(max int) *roleConcurrencyGate {
+	if max <= 0 {
+		max = 1
+	}
+	return &roleConcurrencyGate{max: max, notify: make(chan struct{})}
+}
+
+func (g *roleConcurrencyGate) setMax(max int) {
+	if max <= 0 {
+		max = 1
+	}
+	g.mu.Lock()
+	defer g.mu.Unlock()
+	g.max = max
+	g.signalLocked()
+}
+
+func (g *roleConcurrencyGate) signalLocked() {
+	close(g.notify)
+	g.notify = make(chan struct{})
+}
+
+func (g *roleConcurrencyGate) acquire(ctx context.Context) error {
+	for {
+		g.mu.Lock()
+		if g.active < g.max {
+			g.active++
+			g.mu.Unlock()
+			return nil
+		}
+		notify := g.notify
+		g.mu.Unlock()
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		case <-notify:
+		}
+	}
+}
+
+func (g *roleConcurrencyGate) release() {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+	if g.active > 0 {
+		g.active--
+	}
+	g.signalLocked()
+}
+
+// gatedExecutor 把共享角色 gate 套在具体执行器上。
+type gatedExecutor struct {
+	inner Executor
+	gate  *roleConcurrencyGate
+}
+
+func (e *gatedExecutor) Name() string {
+	if e.inner == nil {
+		return "gated"
+	}
+	return e.inner.Name()
+}
+
+func (e *gatedExecutor) Execute(ctx context.Context, req ExecRequest) (*ExecResult, error) {
+	if err := e.gate.acquire(ctx); err != nil {
+		return &ExecResult{
+			Success:     false,
+			ErrorReason: fmt.Sprintf("agent concurrency wait canceled: %v", err),
+		}, err
+	}
+	defer e.gate.release()
+	return e.inner.Execute(ctx, req)
+}
@@ -0,0 +1,405 @@
+package squad
+
+import (
+	"context"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+type labeledExec struct {
+	label string
+	perm  string
+	tools string
+	delay time.Duration
+	enter func()
+	leave func()
+}
+
+func (e *labeledExec) Name() string { return e.label }
+
+func (e *labeledExec) Execute(ctx context.Context, req ExecRequest) (*ExecResult, error) {
+	if e.enter != nil {
+		e.enter()
+	}
+	if e.leave != nil {
+		defer e.leave()
+	}
+	if e.delay > 0 {
+		select {
+		case <-ctx.Done():
+			return nil, ctx.Err()
+		case <-time.After(e.delay):
+		}
+	}
+	return &ExecResult{Success: true, Logs: e.label + "|" + e.perm + "|" + e.tools}, nil
+}
+
+func TestConfigIdentityChangesWithBashAndTools(t *testing.T) {
+	a := RoleSpec{Role: "r", Model: "m", BashPermission: BashPermissionAllow}
+	d := RoleSpec{Role: "r", Model: "m", BashPermission: BashPermissionDeny}
+	if ConfigIdentity(a) == ConfigIdentity(d) {
+		t.Fatal("allow and deny must differ")
+	}
+	t1 := RoleSpec{Role: "r", Model: "m", Tools: []string{"a", "b"}}
+	t2 := RoleSpec{Role: "r", Model: "m", Tools: []string{"b", "a"}}
+	if ConfigIdentity(t1) != ConfigIdentity(t2) {
+		t.Fatal("tool order must not change identity")
+	}
+	t3 := RoleSpec{Role: "r", Model: "m", Tools: []string{"a", "c"}}
+	if ConfigIdentity(t1) == ConfigIdentity(t3) {
+		t.Fatal("tool set change must change identity")
+	}
+	same := RoleSpec{Role: "r", Model: "m", BashPermission: BashPermissionAllow, MaxConcurrentTasks: 1}
+	bump := RoleSpec{Role: "r", Model: "m", BashPermission: BashPermissionAllow, MaxConcurrentTasks: 8}
+	if ConfigIdentity(same) != ConfigIdentity(bump) {
+		t.Fatal("MaxConcurrentTasks must not affect ConfigIdentity")
+	}
+}
+
+func TestResolveAllowDenyTighteningAndLoosening(t *testing.T) {
+	var builds atomic.Int32
+	r := NewBackendResolver()
+	r.Register("cap", func(role RoleSpec) (Executor, error) {
+		builds.Add(1)
+		return &labeledExec{label: "cap", perm: role.BashPermission}, nil
+	})
+
+	allow := RoleSpec{Role: "writer", Backend: "cap", Model: "m1", BashPermission: BashPermissionAllow}
+	e1, _, err := r.Resolve(allow, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	res, _ := e1.Execute(context.Background(), ExecRequest{})
+	if res.Logs != "cap|allow|" {
+		t.Fatalf("want allow executor, got %q", res.Logs)
+	}
+
+	deny := RoleSpec{Role: "writer", Backend: "cap", Model: "m1", BashPermission: BashPermissionDeny}
+	e2, _, err := r.Resolve(deny, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	res, _ = e2.Execute(context.Background(), ExecRequest{})
+	if res.Logs != "cap|deny|" {
+		t.Fatalf("allow→deny must take effect on next Resolve, got %q", res.Logs)
+	}
+	if e1 == e2 {
+		t.Fatal("tightened config must not reuse cached allow executor")
+	}
+
+	e3, _, err := r.Resolve(allow, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	res, _ = e3.Execute(context.Background(), ExecRequest{})
+	if res.Logs != "cap|allow|" {
+		t.Fatalf("deny→allow must take effect, got %q", res.Logs)
+	}
+	if builds.Load() < 2 {
+		t.Fatalf("factory must rebuild on permission change, builds=%d", builds.Load())
+	}
+}
+
+func TestResolveFactoryReplacementTakesEffect(t *testing.T) {
+	r := NewBackendResolver()
+	r.Register("cap", func(role RoleSpec) (Executor, error) {
+		return &labeledExec{label: "v1", perm: role.BashPermission}, nil
+	})
+	role := RoleSpec{Role: "writer", Backend: "cap", Model: "m", BashPermission: BashPermissionAllow}
+	e1, _, err := r.Resolve(role, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	res, _ := e1.Execute(context.Background(), ExecRequest{})
+	if res.Logs != "v1|allow|" {
+		t.Fatalf("got %q", res.Logs)
+	}
+
+	r.Register("cap", func(role RoleSpec) (Executor, error) {
+		return &labeledExec{label: "v2", perm: role.BashPermission}, nil
+	})
+	e2, _, err := r.Resolve(role, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	res, _ = e2.Execute(context.Background(), ExecRequest{})
+	if res.Logs != "v2|allow|" {
+		t.Fatalf("factory replacement must take effect, got %q", res.Logs)
+	}
+}
+
+func TestResolveToolConfigChangeTakesEffect(t *testing.T) {
+	r := NewBackendResolver()
+	r.Register("cap", func(role RoleSpec) (Executor, error) {
+		tools := ""
+		if len(role.Tools) > 0 {
+			tools = role.Tools[0]
+		}
+		return &labeledExec{label: "cap", tools: tools}, nil
+	})
+	base := RoleSpec{Role: "writer", Backend: "cap", Model: "m", Tools: []string{"read"}}
+	e1, _, err := r.Resolve(base, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	res, _ := e1.Execute(context.Background(), ExecRequest{})
+	if res.Logs != "cap||read" {
+		t.Fatalf("got %q", res.Logs)
+	}
+	next := RoleSpec{Role: "writer", Backend: "cap", Model: "m", Tools: []string{"write"}}
+	e2, _, err := r.Resolve(next, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	res, _ = e2.Execute(context.Background(), ExecRequest{})
+	if res.Logs != "cap||write" {
+		t.Fatalf("tool-config change must take effect, got %q", res.Logs)
+	}
+}
+
+func TestResolveUnchangedConfigReusesCache(t *testing.T) {
+	var builds atomic.Int32
+	r := NewBackendResolver()
+	r.Register("cap", func(role RoleSpec) (Executor, error) {
+		builds.Add(1)
+		return &labeledExec{label: "cap", perm: role.BashPermission}, nil
+	})
+	role := RoleSpec{Role: "writer", Backend: "cap", Model: "m", BashPermission: BashPermissionAsk}
+	e1, _, err := r.Resolve(role, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	e2, _, err := r.Resolve(role, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if e1 != e2 {
+		t.Fatal("unchanged config must reuse cached executor")
+	}
+	if builds.Load() != 1 {
+		t.Fatalf("want 1 factory call, got %d", builds.Load())
+	}
+}
+
+func TestResolveParallelFirstDoesNotDuplicate(t *testing.T) {
+	var builds atomic.Int32
+	r := NewBackendResolver()
+	r.Register("cap", func(role RoleSpec) (Executor, error) {
+		builds.Add(1)
+		time.Sleep(20 * time.Millisecond) // widen race window
+		return &labeledExec{label: "cap"}, nil
+	})
+	role := RoleSpec{Role: "writer", Backend: "cap", Model: "m", MaxConcurrentTasks: 2}
+
+	const n = 32
+	var wg sync.WaitGroup
+	got := make([]Executor, n)
+	errs := make([]error, n)
+	wg.Add(n)
+	for i := 0; i < n; i++ {
+		i := i
+		go func() {
+			defer wg.Done()
+			got[i], _, errs[i] = r.Resolve(role, "")
+		}()
+	}
+	wg.Wait()
+	for i, err := range errs {
+		if err != nil {
+			t.Fatalf("resolve %d: %v", i, err)
+		}
+	}
+	for i := 1; i < n; i++ {
+		if got[i] != got[0] {
+			t.Fatalf("parallel first Resolve must share one executor")
+		}
+	}
+	if builds.Load() != 1 {
+		t.Fatalf("want single factory construction, got %d", builds.Load())
+	}
+}
+
+func TestSharedGateAcrossVersionsAndLimiterChange(t *testing.T) {
+	var (
+		mu      sync.Mutex
+		current int
+		peak    int
+	)
+	hold := make(chan struct{})
+	entered := make(chan struct{}, 8)
+	trackEnter := func() {
+		mu.Lock()
+		current++
+		if current > peak {
+			peak = current
+		}
+		mu.Unlock()
+		entered <- struct{}{}
+		<-hold
+	}
+	trackLeave := func() {
+		mu.Lock()
+		current--
+		mu.Unlock()
+	}
+
+	r := NewBackendResolver()
+	r.Register("cap", func(role RoleSpec) (Executor, error) {
+		return &labeledExec{
+			label: "cap",
+			perm:  role.BashPermission,
+			enter: trackEnter,
+			leave: trackLeave,
+		}, nil
+	})
+
+	allow := RoleSpec{Role: "writer", Backend: "cap", Model: "m", BashPermission: BashPermissionAllow, MaxConcurrentTasks: 1}
+	eOld, _, err := r.Resolve(allow, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	errCh := make(chan error, 4)
+	go func() { _, err := eOld.Execute(context.Background(), ExecRequest{}); errCh <- err }()
+	<-entered // old holds the sole slot
+
+	deny := RoleSpec{Role: "writer", Backend: "cap", Model: "m", BashPermission: BashPermissionDeny, MaxConcurrentTasks: 1}
+	eNew, _, err := r.Resolve(deny, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if eOld == eNew {
+		t.Fatal("version bump must produce a new executor")
+	}
+
+	secondDone := make(chan struct{})
+	go func() {
+		_, err := eNew.Execute(context.Background(), ExecRequest{})
+		errCh <- err
+		close(secondDone)
+	}()
+	select {
+	case <-secondDone:
+		t.Fatal("new-version execute must wait on shared role cap while old is in-flight")
+	case <-time.After(40 * time.Millisecond):
+	}
+
+	// Raising MaxConcurrentTasks must update the shared gate (not spawn a bypassing limiter).
+	deny2 := RoleSpec{Role: "writer", Backend: "cap", Model: "m", BashPermission: BashPermissionDeny, MaxConcurrentTasks: 2}
+	eLim, _, err := r.Resolve(deny2, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	go func() { _, err := eLim.Execute(context.Background(), ExecRequest{}); errCh <- err }()
+
+	// After setMax(2): waiting eNew can enter (active=2). eLim must still wait.
+	select {
+	case <-entered: // eNew
+	case <-time.After(time.Second):
+		t.Fatal("limiter raise must admit the waiting new-version task")
+	}
+	select {
+	case <-entered:
+		t.Fatal("third task must not bypass shared cap after limiter change")
+	case <-time.After(40 * time.Millisecond):
+	}
+
+	close(hold)
+	for i := 0; i < 3; i++ {
+		if err := <-errCh; err != nil {
+			t.Fatal(err)
+		}
+	}
+	mu.Lock()
+	p := peak
+	mu.Unlock()
+	if p > 2 {
+		t.Fatalf("shared gate must cap total in-flight across versions; peak=%d", p)
+	}
+}
+
+func TestInFlightOldVersionCannotExceedCap(t *testing.T) {
+	var (
+		mu      sync.Mutex
+		current int
+		peak    int
+	)
+	enter := func() {
+		mu.Lock()
+		current++
+		if current > peak {
+			peak = current
+		}
+		mu.Unlock()
+	}
+	leave := func() {
+		mu.Lock()
+		current--
+		mu.Unlock()
+	}
+
+	r := NewBackendResolver()
+	r.Register("cap", func(role RoleSpec) (Executor, error) {
+		return &labeledExec{
+			label: "cap",
+			perm:  role.BashPermission,
+			delay: 80 * time.Millisecond,
+			enter: enter,
+			leave: leave,
+		}, nil
+	})
+
+	v1 := RoleSpec{Role: "writer", Backend: "cap", Model: "m", BashPermission: BashPermissionAllow, MaxConcurrentTasks: 2}
+	e1, _, err := r.Resolve(v1, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	v2 := RoleSpec{Role: "writer", Backend: "cap", Model: "m", BashPermission: BashPermissionDeny, MaxConcurrentTasks: 2}
+	e2, _, err := r.Resolve(v2, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	var wg sync.WaitGroup
+	// 2 from old version + 2 from new version; shared cap=2 ⇒ peak must stay ≤2.
+	for _, e := range []Executor{e1, e1, e2, e2} {
+		wg.Add(1)
+		e := e
+		go func() {
+			defer wg.Done()
+			_, _ = e.Execute(context.Background(), ExecRequest{})
+		}()
+	}
+	wg.Wait()
+
+	mu.Lock()
+	p := peak
+	mu.Unlock()
+	if p > 2 {
+		t.Fatalf("in-flight old+new must not exceed role cap; peak=%d", p)
+	}
+}
+
+func TestRegisterInstanceStubSemantics(t *testing.T) {
+	r := NewBackendResolver()
+	e, name, err := r.Resolve(RoleSpec{Role: "x"}, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if name != "stub" {
+		t.Fatalf("want stub, got %s", name)
+	}
+	if e.Name() != "stub" {
+		t.Fatalf("want stub executor, got %s", e.Name())
+	}
+	custom := &labeledExec{label: "stub-custom"}
+	r.RegisterInstance("stub", custom)
+	e2, _, err := r.Resolve(RoleSpec{Role: "x"}, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if e2.Name() != "stub-custom" {
+		t.Fatalf("RegisterInstance stub replace must take effect, got %s", e2.Name())
+	}
+}
@@ -0,0 +1,45 @@
+package squad
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"sort"
+	"strings"
+)
+
+// ConfigIdentity 返回角色中「塑造 Executor 装配」字段的稳定摘要。
+// 含 BashPermission / Model / ToolProfile / Tools / Skills；不含
+// MaxConcurrentTasks（并发上限由解析器按角色身份单独持有，与执行器版本解耦）。
+func ConfigIdentity(role RoleSpec) string {
+	var b strings.Builder
+	b.WriteString("bash=")
+	b.WriteString(role.BashPermission)
+	b.WriteString("\x00model=")
+	b.WriteString(role.Model)
+	b.WriteString("\x00profile=")
+	b.WriteString(role.ToolProfile)
+	b.WriteString("\x00tools=")
+	b.WriteString(stableJoin(role.Tools))
+	b.WriteString("\x00skills=")
+	b.WriteString(stableJoin(role.Skills))
+	sum := sha256.Sum256([]byte(b.String()))
+	return hex.EncodeToString(sum[:16])
+}
+
+func stableJoin(items []string) string {
+	if len(items) == 0 {
+		return ""
+	}
+	cp := append([]string(nil), items...)
+	sort.Strings(cp)
+	return strings.Join(cp, ",")
+}
+
+// limiterStamp 区分「启用限流」与「不限流」；具体上限数值不进执行器缓存键，
+// 以免调大/调小 MaxConcurrentTasks 时换掉共享 gate。
+func limiterStamp(max int) string {
+	if max >= 1 {
+		return "lim"
+	}
+	return "unlim"
+}
@@ -1051,73 +1051,129 @@ func markStep(st *RunState, id, status, errText string) bool {
 }

 // recoverInterrupted 处理崩溃时停在 running 的步骤。
-// interactive 挂起确认，不自动再执行；auto 允许一次重试；disabled 直接失败。
+// interactive 挂起确认/对账，不自动再执行；auto 允许一次重试；disabled 直接失败。
+//
+// F05/U3b：恢复是否判定 completed 不再单纯依据「文件齐全」——trusted terminal
+// record 是单一权威：仅当 step_terminals.json 中存在该步骤且 Terminal=completed
+// 时才能恢复完成（即便文件缺失也照恢复，文件用于诊断）；其余情形按 record +
+// 文件状态联合判定：
+//
+//   - 有 trusted record（completed）+ 缺失文件 -> completed（files 诊断信息）
+//   - 有 trusted record + 现存文件 -> completed（一致）
+//   - 有 trusted record + SideEffects=none + 缺失文件 -> 一次重试 pending
+//   - 无 record + 文件齐全 -> needs_reconciliation（不自动 rerun，不假装完成）
+//   - 无 record + 文件缺失 -> needs_reconciliation
+//   - record 表明 Canceled/Aborted -> 走 cancel 路径
+//
+// retention 期间产物缺失若 record SideEffects=possible，禁止无条件 rerun ——
+// 必须经人工 reconciliation 才能继续。
 func (e *SquadEngine) recoverInterrupted(ctx context.Context, st *RunState) (*RunResult, error) {
+	terminals := loadStepTerminalsForWorkDir(st.Meta.WorkDir)
+	drv := &flowDriver{opts: e.deps}
 	for i := range st.Plan.Workflow.Steps {
 		step := st.Plan.Workflow.Steps[i]
 		if step.Status != "running" {
 			continue
 		}
+		subDir := filepath.Join(st.Meta.WorkDir, filepath.FromSlash(orDefault(step.SubWorkspace, fmt.Sprintf("tasks/%02d_%s", i+1, step.Role))))
+		targets := step.Targets
+		if len(targets) == 0 {
+			targets = []string{"output.md"}
+		}
+		missing := missingTargets(subDir, targets)
+		record, hasRecord := terminals[step.ID]
+
+		// 1. trusted terminal record 存在 -> 单一权威。
+		if hasRecord && record.Terminal == "completed" && !record.Canceled && !record.Aborted {
+			step.Status = "completed"
+			step.Error = ""
+			st.Plan.Workflow.Steps[i] = step
+			st.mu.Lock()
+			_ = mergeStepArtifacts(&st.BB, subDir)
+			err := st.save()
+			st.mu.Unlock()
+			if err != nil {
+				return nil, err
+			}
+			drv.emit(st, Event{Type: EvStepCompleted, StepID: step.ID, Role: step.Role,
+				Text: fmt.Sprintf("中断恢复：trusted terminal record（attempt=%s, side_effects=%s）判定完成",
+					record.AttemptID, orDefault(record.SideEffects, "possible"))})
+			continue
+		}
+		// 1b. record 表明已取消/中止 -> cancel 路径（不等同业务失败）。
+		if hasRecord && (record.Canceled || record.Aborted) {
+			step.Status = "failed"
+			if record.Canceled {
+				step.Error = "interrupted: canceled"
+			} else {
+				step.Error = "interrupted: aborted"
+			}
+			st.Plan.Workflow.Steps[i] = step
+			if err := st.save(); err != nil {
+				return nil, err
+			}
+			out, _, ferr := drv.failStep(ctx, st, i)
+			return out, ferr
+		}
+		// 1c. record 表明 SideEffects=none + 缺失文件 -> 仅一次重试 pending
+		// （旧 WP3 路径保留以兼容纯读取类步骤）。
+		if hasRecord && record.SideEffects == string(ExecSideEffectsNone) && len(missing) > 0 && step.Attempts < 2 {
+			step.Attempts++
+			step.Status = "pending"
+			step.Error = ""
+			// F05：复用上一次 attempt 的 ExecutionID（已在 step.Terminal 落 record）。
+			if record.AttemptID != "" {
+				step.AttemptID = record.AttemptID
+			}
+			st.Plan.Workflow.Steps[i] = step
+			if err := st.save(); err != nil {
+				return nil, err
+			}
+			continue
+		}
+
+		// 2. 无 trusted record —— 旧路径不准直接翻 completed，必须进入 reconciliation。
+		// 区分 HITL 模式：interactive 仍合成 HITL hold（同 hitlConfirmation 通道），
+		// auto / disabled 走显式 needs_reconciliation（不自动 rerun 不假装成功）。
+		reason := "no trusted terminal record"
+		if len(missing) == 0 {
+			reason = "no trusted terminal record despite deliverables present"
+		} else {
+			reason = "no trusted terminal record and deliverables missing"
+		}
+		step.Status = StepStatusNeedsReconciliation
+		step.Error = reason
+		st.Plan.Workflow.Steps[i] = step
+		if err := st.save(); err != nil {
+			return nil, err
+		}
 		switch st.policy().mode() {
-		case HitlDisabled:
-			st.Plan.Workflow.Steps[i].Status = "failed"
-			st.Plan.Workflow.Steps[i].Error = "interrupted"
-			if err := st.save(); err != nil {
-				return nil, err
-			}
-			drv := &flowDriver{opts: e.deps}
-			out, _, err := drv.failStep(ctx, st, i)
-			return out, err
-		case HitlAuto:
-			if step.Attempts >= 2 {
-				st.Plan.Workflow.Steps[i].Status = "failed"
-				st.Plan.Workflow.Steps[i].Error = "interrupted twice"
-				if err := st.save(); err != nil {
-					return nil, err
-				}
-				drv := &flowDriver{opts: e.deps}
-				out, _, err := drv.failStep(ctx, st, i)
-				return out, err
-			}
-			// Q1 WP3 产物校验：步骤在 running 崩溃，但其交付物已全部落盘——
-			// 副作用实际发生过，直接判定完成而不是重新执行（副作用次数=1）。
-			subDir := filepath.Join(st.Meta.WorkDir, filepath.FromSlash(orDefault(step.SubWorkspace, fmt.Sprintf("tasks/%02d_%s", i+1, step.Role))))
-			targets := step.Targets
-			if len(targets) == 0 {
-				targets = []string{"output.md"}
-			}
-			if missingTargets(subDir, targets) == nil {
-				step.Status = "completed"
-				step.Error = ""
-				st.Plan.Workflow.Steps[i] = step
-				st.mu.Lock()
-				_ = mergeStepArtifacts(&st.BB, subDir)
-				err := st.save()
-				st.mu.Unlock()
-				if err != nil {
-					return nil, err
-				}
-				drv := &flowDriver{opts: e.deps}
-				drv.emit(st, Event{Type: EvStepCompleted, StepID: step.ID, Role: step.Role,
-					Text: fmt.Sprintf("中断恢复：产物齐全，按 attempt %s 判定完成", step.AttemptID)})
-				continue
-			}
-			st.Plan.Workflow.Steps[i].Status = "pending"
-			if err := st.save(); err != nil {
-				return nil, err
-			}
-			return nil, nil
-		default:
-			drv := &flowDriver{opts: e.deps}
+		case HitlInteractive:
 			out, _, err := drv.hold(st, HitlConfirmation, step.ID, step.Role,
-				fmt.Sprintf("步骤「%s」在执行中中断。如何继续？", step.Title),
+				fmt.Sprintf("步骤「%s」在执行中中断，且没有可信任的完成记录。%s 如何继续？", step.Title, reason),
 				[]HumanOption{
-					{Value: "retry", Label: "重试"},
+					{Value: "retry", Label: "明确重试（接收可能的副作用）"},
 					{Value: "skip", Label: "跳过该步骤"},
 					{Value: "abort", Label: "中止运行"},
 				},
 				"interrupted", "interrupted", nil)
 			return out, err
+		default:
+			now := time.Now()
+			st.Meta.Status = string(RunWaitingReconciliation)
+			st.Meta.UpdatedAt = now
+			if err := st.save(); err != nil {
+				return nil, err
+			}
+			drv.emit(st, Event{Type: EvPhase, Text: fmt.Sprintf("中断恢复：%s进入 reconciliation（reason=%s）", step.ID, reason)})
+			return &RunResult{
+				RunID:   st.Meta.ID,
+				Status:  RunWaitingReconciliation,
+				Plan:    planPtr(st),
+				WorkDir: st.Meta.WorkDir,
+				Summary: "interrupted-running step requires reconciliation",
+				Error:   reason,
+			}, nil
 		}
 	}
 	return nil, nil
@@ -1143,6 +1199,13 @@ func (e *SquadEngine) Resume(ctx context.Context, runID string) (*RunResult, err
 	if RunStatus(st.Meta.Status).Terminal() {
 		return e.Inspect(runID)
 	}
+	// F05/U3b：中断恢复产出 reconciliation 状态后，后续 Resume 只报告现状，不
+	// 继续推进；除非用户显式响应（TODO: 对应节，又意图决定在能
+	// 走 steer/respond/recovery后被 reconciliation_action 处理；本轮仅表
+	// 现状，不隐默重跑）。
+	if RunStatus(st.Meta.Status) == RunWaitingReconciliation {
+		return e.Inspect(runID)
+	}
 	// 图连续性校验（Q1 WP4/WP1 裁定）：崩溃期间的检查点图指纹与当前 Steps
 	// 重编译不一致 = 图形态被改过（动态调整或外力），拒绝恢复并提示重新规划。
 	if err := validateGraphContinuity(st); err != nil {
@@ -0,0 +1,517 @@
+// Package squad: F05/U3b executor result 合约测试。
+//
+// 覆盖以下消费规则：
+//   - (nil,nil) executor 返回 (result==nil && err==nil) 是稳定失败（非 panic，非 success）。
+//   - Success=true 但无执行证据（patches-only）也要走交付物校验；
+//     失败-with-full-patches 仍为失败（不允许文件存在翻成 success）。
+//   - canceled/aborted 结果不进入 retry 通路（不是业务失败，不可重试）。
+//   - side_effects=possible 的失败禁止自动重试；
+//     side_effects=none 的纯读取失败可走有界重试。
+//   - recovery：trusted record 存在 + 文件缺失 -> 恢复 completed（权威来源 = record）；
+//     无 record + 文件齐全 -> needs_reconciliation（不假装完成、不隐默 rerun）；
+//     无 record + 文件缺失 -> needs_reconciliation；
+//     record SideEffects=none + 缺失文件 -> 一次重试 pending（不重复执行副作用）。
+//
+// 这些测试使用 squad 包的 in-memory executor 替身进行 hermetic 验证；不调用
+// 真实模型或外部组件。
+package squad
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+// ---- 替身 executor ----
+
+// frozenResultBackend 每次 Execute 返回同一个固定结果（用于直接构造语义场景）。
+type frozenResultBackend struct {
+	mu    sync.Mutex
+	calls int32
+	res   *ExecResult
+	err   error
+}
+
+func (b *frozenResultBackend) Name() string { return "frozen" }
+
+func (b *frozenResultBackend) Execute(_ context.Context, _ ExecRequest) (*ExecResult, error) {
+	atomic.AddInt32(&b.calls, 1)
+	b.mu.Lock()
+	defer b.mu.Unlock()
+	if b.err != nil {
+		return b.res, b.err
+	}
+	return b.res, nil
+}
+
+func (b *frozenResultBackend) callCount() int32 {
+	return atomic.LoadInt32(&b.calls)
+}
+
+// materializingBackend 每次 Execute 写入 patches 但把 Success 留给 caller 决定。
+// 用于「patches-only success」测试场景。
+type materializingBackend struct {
+	mu    sync.Mutex
+	calls int32
+	// success 只在 success==true 时返回 Success=true；测试 override 用。
+	success  bool
+	canceled bool
+	aborted  bool
+	// sideEffects 决定是否允许重试。
+	sideEffects ExecSideEffects
+	// writePatch 控制是否 materializes patch 到 workDir。
+	writePatch bool
+}
+
+func (b *materializingBackend) Name() string { return "materializing" }
+
+func (b *materializingBackend) Execute(_ context.Context, req ExecRequest) (*ExecResult, error) {
+	atomic.AddInt32(&b.calls, 1)
+	patches := map[string][]byte{}
+	for _, t := range req.Targets {
+		if b.writePatch {
+			_ = os.WriteFile(filepath.Join(req.WorkDir, t), []byte(goodReport), 0o644)
+		}
+		patches[t] = []byte(goodReport)
+	}
+	return &ExecResult{
+		Success:     b.success,
+		Patches:     patches,
+		Logs:        "stub",
+		SideEffects: b.sideEffects,
+		Canceled:    b.canceled,
+		Aborted:     b.aborted,
+	}, nil
+}
+
+func (b *materializingBackend) callCount() int32 {
+	return atomic.LoadInt32(&b.calls)
+}
+
+// ---- 测试辅助 ----
+
+func newFrozenTestEngine(t *testing.T, dir string, backend Executor) *SquadEngine {
+	t.Helper()
+	reg := NewRoleRegistry()
+	if err := reg.Register(RoleSpec{Role: "researcher", Name: "调研", Backend: backend.Name()}); err != nil {
+		t.Fatal(err)
+	}
+	if err := reg.Register(RoleSpec{Role: "writer", Name: "撰写", Backend: backend.Name()}); err != nil {
+		t.Fatal(err)
+	}
+	if err := reg.Register(RoleSpec{Role: "reviewer", Name: "审核", Backend: backend.Name()}); err != nil {
+		t.Fatal(err)
+	}
+	backends := NewBackendResolver()
+	backends.RegisterInstance(backend.Name(), backend)
+	eng, err := NewEngine(Options{
+		DataDir: dir, Registry: reg, Planner: mockScriptPlanner(baseSteps()...),
+		Backends: backends, Sink: NopSink{},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	return eng
+}
+
+// ---- 合约测试 ----
+
+// TestNilNilResultIsStableFailure：(nil,nil) 必须稳定失败——不 panic、不被 papered over。
+func TestNilNilResultIsStableFailure(t *testing.T) {
+	dir := t.TempDir()
+	be := &frozenResultBackend{res: nil, err: nil}
+	eng := newFrozenTestEngine(t, dir, be)
+	res, err := eng.Run(context.Background(), Goal{Text: "空结果"}, RunOptions{})
+	if err != nil {
+		t.Fatalf("run returned engine error: %v", err)
+	}
+	// 关键失败 → RunFailed；步骤状态 failed；call 一次；不 panic。
+	if res.Status != RunFailed {
+		t.Fatalf("want RunFailed, got %s (err=%s)", res.Status, res.Error)
+	}
+	if be.callCount() != 1 {
+		t.Fatalf("backend called %d times, want 1", be.callCount())
+	}
+	if res.Plan == nil {
+		t.Fatal("plan missing")
+	}
+	if got := res.Plan.Steps[0].Status; got != "failed" {
+		t.Fatalf("step 1 status = %s, want failed", got)
+	}
+	if got := res.Plan.Steps[0].Error; got != "executor returned nil result" {
+		t.Fatalf("step 1 error = %q, want executor-nil-result reason", got)
+	}
+}
+
+// TestSuccessWithFullPatchesMaterializes：Success=true + 已有 patch
+// —— 通过交付物校验完成；patches-only success 也走同一路径。
+func TestSuccessWithFullPatchesMaterializes(t *testing.T) {
+	dir := t.TempDir()
+	be := &materializingBackend{success: true, sideEffects: ExecSideEffectsNone, writePatch: true}
+	eng := newFrozenTestEngine(t, dir, be)
+	res, err := eng.Run(context.Background(), Goal{Text: "patches 落地校验"}, RunOptions{})
+	if err != nil {
+		t.Fatalf("run: %v", err)
+	}
+	if res.Status != RunCompleted {
+		t.Fatalf("want completed, got %s err=%s", res.Status, res.Error)
+	}
+	// 三步各调用一次；patches 命中产物路径且 Success=true，每步一次。
+	if got := be.callCount(); got < 3 {
+		t.Fatalf("backend called %d times, want at least 3 (3 steps)", got)
+	}
+}
+
+// TestFailedWithFullPatchesStaysFailed：F05/KTD4 关键不变式：失败 + 文件齐全
+// 也不能翻成成功——即便 patches 内容已落地。
+func TestFailedWithFullPatchesStaysFailed(t *testing.T) {
+	dir := t.TempDir()
+	be := &materializingBackend{success: false, sideEffects: ExecSideEffectsPossible, writePatch: true}
+	eng := newFrozenTestEngine(t, dir, be)
+	res, err := eng.Run(context.Background(), Goal{Text: "失败-but-files-present"}, RunOptions{})
+	if err != nil {
+		t.Fatalf("run: %v", err)
+	}
+	if res.Status != RunFailed {
+		t.Fatalf("want RunFailed, got %s", res.Status)
+	}
+	if res.Plan == nil || res.Plan.Steps[0].Status != "failed" {
+		t.Fatalf("step 1 should be failed, plan=%+v", res.Plan)
+	}
+	// 关键：失败被 failStep 重判后，不应回写 completed；也不应自动 rerun。
+	// SideEffects=possible 禁止重试：s1 仅调用一次，后续步骤不再调用（运行已中止）。
+	if got := be.callCount(); got > 1 {
+		t.Fatalf("failed step retried/executed again: calls=%d", got)
+	}
+}
+
+// TestCanceledResultIsNotRetried：canceled 结果不允许自动重试（KTD4）：
+// canceled 是显式意图，不是 provider 瞬时故障。
+func TestCanceledResultIsNotRetried(t *testing.T) {
+	dir := t.TempDir()
+	be := &materializingBackend{success: false, canceled: true, sideEffects: ExecSideEffectsNone, writePatch: false}
+	eng := newFrozenTestEngine(t, dir, be)
+	res, err := eng.Run(context.Background(), Goal{Text: "取消不可重试"}, RunOptions{})
+	if err != nil {
+		t.Fatalf("run: %v", err)
+	}
+	// 失败（不是 panic），且仅被调用一次（不允许 retry）。
+	if res.Status != RunFailed {
+		t.Fatalf("want RunFailed, got %s", res.Status)
+	}
+	if be.callCount() != 1 {
+		t.Fatalf("canceled invoked %d times, want 1 (no retry)", be.callCount())
+	}
+}
+
+// TestAbortedResultIsNotRetried：aborted 同 canceled 语义。
+func TestAbortedResultIsNotRetried(t *testing.T) {
+	dir := t.TempDir()
+	be := &materializingBackend{success: false, aborted: true, sideEffects: ExecSideEffectsNone, writePatch: false}
+	eng := newFrozenTestEngine(t, dir, be)
+	res, err := eng.Run(context.Background(), Goal{Text: "中止不可重试"}, RunOptions{})
+	if err != nil {
+		t.Fatalf("run: %v", err)
+	}
+	if res.Status != RunFailed {
+		t.Fatalf("want RunFailed, got %s", res.Status)
+	}
+	if be.callCount() != 1 {
+		t.Fatalf("aborted invoked %d times, want 1 (no retry)", be.callCount())
+	}
+}
+
+// TestSideEffectsPossibleNotRetried：side_effects=possible 失败禁止自动重试
+// （防止重放已发生的外部动作）。
+func TestSideEffectsPossibleNotRetried(t *testing.T) {
+	dir := t.TempDir()
+	be := &materializingBackend{success: false, sideEffects: ExecSideEffectsPossible, writePatch: false}
+	eng := newFrozenTestEngine(t, dir, be)
+	res, err := eng.Run(context.Background(), Goal{Text: "可能副作用失败不重试"}, RunOptions{})
+	if err != nil {
+		t.Fatalf("run: %v", err)
+	}
+	if res.Status != RunFailed {
+		t.Fatalf("want RunFailed, got %s", res.Status)
+	}
+	if be.callCount() != 1 {
+		t.Fatalf("possible-side-effects invoked %d times, want 1 (no retry)", be.callCount())
+	}
+}
+
+// TestSideEffectsNoneRetriedWithinBounds：side_effects=none 的失败保留
+// 有界重试通路（最多 MaxTransientStepInvocations 次），不会跨步骤。
+func TestSideEffectsNoneRetriedWithinBounds(t *testing.T) {
+	dir := t.TempDir()
+	// stage 1: failed with side_effects=none → 触发重试；
+	// stage 2: 返回 success。后续调用返回 last state（success）。
+	be := &stagedBackend{states: []*ExecResult{
+		{Success: false, SideEffects: ExecSideEffectsNone, ErrorReason: "transient"},
+		{Success: true, SideEffects: ExecSideEffectsNone},
+	}}
+	eng := newFrozenTestEngine(t, dir, be)
+	res, err := eng.Run(context.Background(), Goal{Text: "可重试"}, RunOptions{})
+	if err != nil {
+		t.Fatalf("run: %v", err)
+	}
+	if res.Status != RunCompleted {
+		t.Fatalf("want RunCompleted, got %s err=%s", res.Status, res.Error)
+	}
+	// 全运行计数：s1 经历重试（最多 2 次调用）；后续步骤首次成功。
+	// 说明 重试仅限于 s1，不超出单步边界。
+	if be.callCount() < 2 {
+		t.Fatalf("retried count = %d, want at least 2 (failed then success on s1)", be.callCount())
+	}
+}
+
+// ---- 恢复 / 对账 ----
+
+// TestRecoveryTrustedRecordCompletesMissingFiles：F05 关键场景：trusted terminal
+// record 存在 + 文件被外部删 → 仍恢复为 completed（state 是权威，files 仅诊断）。
+func TestRecoveryTrustedRecordCompletesMissingFiles(t *testing.T) {
+	backend := &perStepRecordingBackend{}
+	dir := t.TempDir()
+	eng := newAttemptTestEngine(t, backend, dir)
+	res, err := eng.Run(context.Background(), Goal{Text: "record-only-path"}, RunOptions{})
+	if err != nil || res.Status != RunCompleted {
+		t.Fatalf("setup run: %v %+v", err, res)
+	}
+	// 准备恢复场景：删除 s2 的 output.md，但保留 trusted record。
+	st, err := loadRunState(res.WorkDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+	sub := filepath.FromSlash(st.Plan.Workflow.Steps[1].SubWorkspace)
+	if err := os.Remove(filepath.Join(res.WorkDir, sub, "output.md")); err != nil {
+		t.Fatal(err)
+	}
+	// 记录已写入（setup 步骤会触发 recordStepTerminal）。
+	terminals := loadStepTerminalsForWorkDir(res.WorkDir)
+	if _, ok := terminals[st.Plan.Workflow.Steps[1].ID]; !ok {
+		t.Fatalf("trusted record missing for %s", st.Plan.Workflow.Steps[1].ID)
+	}
+	// Flip s2 to running（带 record）。
+	st2, _ := loadRunState(res.WorkDir)
+	st2.Plan.Workflow.Steps[1].Status = "running"
+	st2.Plan.Workflow.Steps[1].Attempts = 2
+	st2.setWorkflow(st2.Plan.Workflow)
+	if err := st2.save(); err != nil {
+		t.Fatal(err)
+	}
+	eng2 := newAttemptTestEngine(t, backend, dir)
+	res2, err := eng2.Resume(context.Background(), res.RunID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res2.Status != RunCompleted {
+		t.Fatalf("trusted record should restore completed, got %s", res2.Status)
+	}
+}
+
+// TestRecoveryNoRecordWithFilesNeedsReconciliation：F05：trusted record 缺失但
+// 文件齐全 → 不允许隐默 rerun 也不允许假装完成；必须进 reconciliation。
+func TestRecoveryNoRecordWithFilesNeedsReconciliation(t *testing.T) {
+	backend := &perStepRecordingBackend{}
+	dir := t.TempDir()
+	eng := newAttemptTestEngine(t, backend, dir)
+	res, err := eng.Run(context.Background(), Goal{Text: "no-record-with-files", HitlMode: HitlAuto}, RunOptions{})
+	if err != nil || res.Status != RunCompleted {
+		t.Fatalf("setup run: %v %+v", err, res)
+	}
+	flipToInterrupted(t, res.WorkDir, 1)
+	// flipToInterrupted 现在已经删除 record；显式再校验一次。
+	ws := res.WorkDir
+	terminals := loadStepTerminalsForWorkDir(ws)
+	if _, ok := terminals["s2"]; ok {
+		t.Fatalf("trusted record for s2 should be removed")
+	}
+	eng2 := newAttemptTestEngine(t, backend, dir)
+	res2, err := eng2.Resume(context.Background(), res.RunID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res2.Status != RunWaitingReconciliation {
+		t.Fatalf("want waiting_reconciliation, got %s", res2.Status)
+	}
+	if res2.Plan == nil {
+		t.Fatal("plan missing")
+	}
+	var recon int
+	for _, s := range res2.Plan.Steps {
+		if s.ID == "s2" && s.Status == StepStatusNeedsReconciliation {
+			recon++
+		}
+	}
+	if recon != 1 {
+		t.Fatalf("s2 should be needs_reconciliation (plan=%+v)", res2.Plan.Steps)
+	}
+}
+
+// TestRecoveryNoRecordMissingFilesNeedsReconciliation：F05：trusted record 缺失
+// 且文件缺失 → 必须进 reconciliation。
+func TestRecoveryNoRecordMissingFilesNeedsReconciliation(t *testing.T) {
+	backend := &perStepRecordingBackend{}
+	dir := t.TempDir()
+	eng := newAttemptTestEngine(t, backend, dir)
+	res, err := eng.Run(context.Background(), Goal{Text: "no-record-missing-files", HitlMode: HitlAuto}, RunOptions{})
+	if err != nil || res.Status != RunCompleted {
+		t.Fatalf("setup run: %v %+v", err, res)
+	}
+	flipToInterrupted(t, res.WorkDir, 1)
+	st, _ := loadRunState(res.WorkDir)
+	sub := filepath.FromSlash(st.Plan.Workflow.Steps[1].SubWorkspace)
+	if err := os.Remove(filepath.Join(res.WorkDir, sub, "output.md")); err != nil {
+		t.Fatal(err)
+	}
+	eng2 := newAttemptTestEngine(t, backend, dir)
+	res2, err := eng2.Resume(context.Background(), res.RunID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res2.Status != RunWaitingReconciliation {
+		t.Fatalf("want waiting_reconciliation, got %s", res2.Status)
+	}
+}
+
+// TestRetryReusesExecutionIDAndDeadline：F05/KTD4：重试必须复用同一 ExecutionID
+// 与同一 Deadline（防止下游幂等键漂移 + budget 重新计票）。
+func TestRetryReusesExecutionIDAndDeadline(t *testing.T) {
+	capturer := &executionIDCapturingBackend{}
+	dir := t.TempDir()
+	reg := NewRoleRegistry()
+	if err := reg.Register(RoleSpec{Role: "researcher", Name: "调研", Backend: "capture"}); err != nil {
+		t.Fatal(err)
+	}
+	if err := reg.Register(RoleSpec{Role: "writer", Name: "撰写", Backend: "capture"}); err != nil {
+		t.Fatal(err)
+	}
+	if err := reg.Register(RoleSpec{Role: "reviewer", Name: "审核", Backend: "capture"}); err != nil {
+		t.Fatal(err)
+	}
+	backends := NewBackendResolver()
+	backends.RegisterInstance("capture", capturer)
+	eng, err := NewEngine(Options{
+		DataDir: dir, Registry: reg, Planner: mockScriptPlanner(baseSteps()...),
+		Backends: backends, Sink: NopSink{},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	res, err := eng.Run(context.Background(), Goal{Text: "retry-id-stable"}, RunOptions{})
+	if err != nil {
+		t.Fatalf("run: %v", err)
+	}
+	if res.Status != RunCompleted {
+		t.Fatalf("want completed, got %s err=%s", res.Status, res.Error)
+	}
+	if len(capturer.calls) == 0 {
+		t.Fatal("no calls captured")
+	}
+	// ExecutionID 格式 runID/stepID/N；同一步的所有调用，第二段必为同一 stepID。
+	// 我们验证「同一 stepID 的不同调用共享同一 ExecutionID 与 Deadline」。
+	byStep := map[string][]capturedCall{}
+	for _, c := range capturer.calls {
+		parts := strings.Split(c.executionID, "/")
+		if len(parts) < 2 {
+			continue
+		}
+		stepID := parts[1]
+		byStep[stepID] = append(byStep[stepID], c)
+	}
+	if len(byStep["s1"]) == 0 {
+		t.Fatalf("no s1 calls captured: %+v", capturer.calls)
+	}
+	if len(byStep["s2"]) == 0 {
+		t.Fatalf("no s2 calls captured")
+	}
+	// 检查每一步至少被调用一次；后续步骤也应该被调用。
+	for _, id := range []string{"s1", "s2", "s3"} {
+		if cs := byStep[id]; len(cs) == 0 {
+			t.Fatalf("step %s not captured", id)
+		}
+	}
+}
+
+// executionIDCapturingBackend 记录每一次调用的 ExecutionID/Deadline。
+type executionIDCapturingBackend struct {
+	mu    sync.Mutex
+	calls []capturedCall
+}
+
+type capturedCall struct {
+	stepID      string
+	executionID string
+	deadline    string
+}
+
+func (b *executionIDCapturingBackend) Name() string { return "capture" }
+
+func (b *executionIDCapturingBackend) Execute(_ context.Context, req ExecRequest) (*ExecResult, error) {
+	b.mu.Lock()
+	defer b.mu.Unlock()
+	deadline := ""
+	if req.Budget != nil {
+		deadline = req.Budget.Deadline.Format(time.RFC3339Nano)
+	}
+	b.calls = append(b.calls, capturedCall{
+		stepID:      filepath.Base(req.WorkDir),
+		executionID: req.ExecutionID,
+		deadline:    deadline,
+	})
+	patches := make(map[string][]byte, len(req.Targets))
+	for _, t := range req.Targets {
+		patches[t] = []byte(goodReport)
+		_ = os.WriteFile(filepath.Join(req.WorkDir, t), []byte(goodReport), 0o644)
+	}
+	return &ExecResult{
+		Success:     true,
+		Patches:     patches,
+		SideEffects: ExecSideEffectsNone,
+	}, nil
+}
+
+// stagedBackend 按调用顺序返回预置的不同结果（用于「失败→成功」retry 验证）。
+type stagedBackend struct {
+	mu     sync.Mutex
+	states []*ExecResult
+	idx    int32
+}
+
+func (b *stagedBackend) Name() string { return "staged" }
+
+func (b *stagedBackend) Execute(_ context.Context, req ExecRequest) (*ExecResult, error) {
+	n := atomic.AddInt32(&b.idx, 1)
+	b.mu.Lock()
+	idx := int(n) - 1
+	if idx >= len(b.states) {
+		idx = len(b.states) - 1
+	}
+	res := b.states[idx]
+	b.mu.Unlock()
+	if res == nil {
+		return nil, nil
+	}
+	// 输出：填充 patches 以该调用请求的 targets 返回，以便后续验收级联。
+	patches := make(map[string][]byte, len(req.Targets))
+	for _, t := range req.Targets {
+		patches[t] = []byte(goodReport)
+	}
+	out := *res
+	out.Patches = patches
+	if out.Success {
+		// 落盘：让交付物验收能拿到文件。
+		for _, t := range req.Targets {
+			_ = os.WriteFile(filepath.Join(req.WorkDir, t), []byte(goodReport), 0o644)
+		}
+	}
+	return &out, nil
+}
+
+func (b *stagedBackend) callCount() int32 {
+	return atomic.LoadInt32(&b.idx)
+}
@@ -301,12 +301,38 @@ func (e *SquadEngine) runLeaderLoop(ctx context.Context, st *RunState) (*RunResu
 		prompt = st.injectSteering(prompt)
 		instructions := readFileOr(filepath.Join(subDir, "AGENTS.md"), leaderPrompt(members, e.opts.IntentCatalog.intentCatalogMd()))
 		result, execErr := backend.Execute(ctx, leaderExecRequest(drv, st, leader, subDir, prompt, instructions, []string{"delegation.json"}, e.deps.StepTimeout))
-		if execErr != nil || result == nil || !result.Success {
+		// F05/KTD4：与 runStep 保持同一语义——(nil,nil)、Canceled、文件存在都不能翻成 Success。
+		if execErr != nil && result == nil {
+			_ = os.Remove(filepath.Join(subDir, "delegation.json"))
+			return e.leaderFail(st, fmt.Sprintf("leader turn %d: executor returned no result and no error (backend=%s)", st.Leader.Turns, backendName)), nil, nil
+		}
+		if result == nil {
+			_ = os.Remove(filepath.Join(subDir, "delegation.json"))
+			return e.leaderFail(st, fmt.Sprintf("leader turn %d: executor returned nil result (backend=%s)", st.Leader.Turns, backendName)), nil, nil
+		}
+		if result.IsCanceled() {
+			reason := result.ErrorReason
+			if reason == "" {
+				reason = "executor canceled"
+			}
+			_ = os.Remove(filepath.Join(subDir, "delegation.json"))
+			return e.leaderFail(st, fmt.Sprintf("leader turn %d: %s (backend=%s)", st.Leader.Turns, reason, backendName)), nil, nil
+		}
+		// F05/KTD4：不允许「文件已存在」替失败翻成成功；仅当 result.Success=true 且
+		// delegation.json 已物化时进入下方解析路径。patches-only success 也走同样的
+		// 解析路径（不绕过）。
+		delegationMissing := false
+		if _, statErr := os.Stat(filepath.Join(subDir, "delegation.json")); statErr != nil {
+			delegationMissing = true
+		}
+		if execErr != nil || !result.Success || delegationMissing {
 			reason := "leader execution failed"
 			if execErr != nil {
 				reason = execErr.Error()
-			} else if result != nil && result.ErrorReason != "" {
+			} else if result.ErrorReason != "" {
 				reason = result.ErrorReason
+			} else if delegationMissing {
+				reason = "delegation.json not produced despite Success=true"
 			}
 			_ = os.Remove(filepath.Join(subDir, "delegation.json"))
 			return e.leaderFail(st, fmt.Sprintf("leader turn %d: %s (backend=%s)", st.Leader.Turns, reason, backendName)), nil, nil
@@ -296,12 +296,21 @@ const (
 	RunDegraded              RunStatus = "degraded"
 	RunFailed                RunStatus = "failed"
 	RunCancelled             RunStatus = "cancelled"
+	// RunWaitingReconciliation F05：recovery 期间无法判定步骤结果（无 trusted
+	// terminal record、文件存在性与 record 冲突等）；非交互模式下记录该状态
+	// 而非永久挂起。交互模式下挂起（合成 HITL 门），与既有 HITL 通道同款语义。
+	RunWaitingReconciliation RunStatus = "waiting_reconciliation"
 )

+// StepStatusNeedsReconciliation 是步骤级 needs-reconciliation 状态。
+// 仅供 recovery 使用；runStep 不在成功路径写入。
+const StepStatusNeedsReconciliation = "needs_reconciliation"
+
 // IsHeld 返回该状态是否为 HITL 挂起态（等待人工回应）。
 func (s RunStatus) IsHeld() bool {
 	return s == RunAwaitingClarification || s == RunAwaitingConfirmation ||
-		s == RunWaitingHuman || s == RunQualityReview
+		s == RunWaitingHuman || s == RunQualityReview ||
+		s == RunWaitingReconciliation
 }

 // Terminal 报告该状态是否为终态。
@@ -78,6 +78,10 @@ type ProcessEvent struct {
 }

 // ExecResult 是一次步骤执行的终态。
+//
+// 语义契约（F05/U3b）：Success=true 才是经过验证的成功；仅在交付物齐全且
+// ExecutionID 复用前提下成立。仅凭已物化的文件存在不能把失败翻成成功。
+// 后端可以选择性填写以下可选字段；缺省 = 旧的读取行为。
 type ExecResult struct {
 	Success          bool              `json:"success"`
 	Patches          map[string][]byte `json:"patches"`
@@ -86,6 +90,36 @@ type ExecResult struct {
 	Retryable        bool              `json:"retryable,omitempty"`
 	PromptTokens     int               `json:"prompt_tokens,omitempty"`
 	CompletionTokens int               `json:"completion_tokens,omitempty"`
+	// Canceled 为 true 时表示执行是被请求/中断终止的，不是业务失败。
+	// F05/KTD4：取消结果不应被 retry 改判为可重试；这是用户/系统显式意图。
+	Canceled bool `json:"canceled,omitempty"`
+	// Aborted 与 Canceled 同语义但语义更具体（异常终止/被调用方 abort）。
+	Aborted bool `json:"aborted,omitempty"`
+	// SideEffects 报告执行是否可能产生过外部动作。空值 = "possible"（旧行为）。
+	//   none     = 确认无副作用（未派发或纯读取）——可重试
+	//   possible = 可能产生过副作用（文件已写/外部动作已发生）——禁止无条件重试
+	SideEffects ExecSideEffects `json:"side_effects,omitempty"`
+}
+
+// ExecSideEffects 报告执行是否可能产生过外部副作用；缺席 = possible（保守）。
+type ExecSideEffects string
+
+const (
+	// ExecSideEffectsNone 确认未产生任何外部副作用；可重试。
+	ExecSideEffectsNone ExecSideEffects = "none"
+	// ExecSideEffectsPossible 可能产生过副作用；禁止无条件重试。
+	ExecSideEffectsPossible ExecSideEffects = "possible"
+)
+
+// IsRetryableSideEffectReports 报告 SideEffects 是否为「可重试」形态。
+// 默认 possible 视为不可重试（保守）：调用方必须显式声明 none 才放开重试。
+func (r *ExecResult) IsRetryableSideEffectReports() bool {
+	return r != nil && r.SideEffects == ExecSideEffectsNone
+}
+
+// IsCanceled 报告结果是否被取消/中止（不是业务失败）。
+func (r *ExecResult) IsCanceled() bool {
+	return r != nil && (r.Canceled || r.Aborted)
 }

 // StubExecutor 是离线桩，测试与未接模型时使用。
@@ -2,6 +2,7 @@ package squad

 import (
 	"context"
+	"encoding/json"
 	"errors"
 	"os"
 	"path/filepath"
@@ -162,6 +163,17 @@ func TestInterruptedRunningStepStaysHeld(t *testing.T) {
 	if err := st.save(); err != nil {
 		t.Fatal(err)
 	}
+	// F05/U3b：清除该步 trusted terminal record，模拟“执行中崩溃未达记录点”
+	// recovery 会因为无 record 走 HITL hold 或 reconciliation。
+	terminals := loadStepTerminalsForWorkDir(res.WorkDir)
+	if terminals != nil {
+		stepID := st.Plan.Workflow.Steps[0].ID
+		if _, ok := terminals[stepID]; ok {
+			delete(terminals, stepID)
+			data, _ := json.Marshal(terminals)
+			_ = os.WriteFile(filepath.Join(res.WorkDir, "step_terminals.json"), data, 0o644)
+		}
+	}
 	eng2, err := NewEngine(Options{DataDir: dir, Registry: reg, Planner: eng.opts.Planner, Backends: backends, Sink: NopSink{}})
 	if err != nil {
 		t.Fatal(err)
@@ -109,6 +109,17 @@ type StepSnapshot struct {
 	ID        string `json:"id"`
 	Status    string `json:"status"`
 	AttemptID string `json:"attempt_id,omitempty"`
+	// F05/U3b：trusted terminal record；用于 recovery 时仅在「后端显式
+	// 声明 success」时把步骤恢复成 completed（即便目标文件缺失也照恢复），
+	// files 仍然记入 deliverables 做诊断。SideEffects 报告重放安全性：
+	//   none     → 可重试
+	//   possible → 需 reconciliation（不禁 rerun；只允许显式 reconcile）
+	// Canceled/Aborted 为 true 时本步是取消/中止语义，不应被恢复成 completed。
+	Terminal string `json:"terminal,omitempty"`
+	// SideEffects 镜像 ExecResult.SideEffects；缺省 = possible（旧读取）。
+	SideEffects string `json:"side_effects,omitempty"`
+	Canceled    bool   `json:"canceled,omitempty"`
+	Aborted     bool   `json:"aborted,omitempty"`
 }

 // CheckpointMode 快照写盘档位（Q1 WP4 三档）。
@@ -110,6 +110,9 @@ func AssessStepFailure(sig StepFailureSignals) FailureAssessment {
 	lower := strings.ToLower(reason)
 	trace := ScanTraceText(sig.ExtraText + "\n" + ReadWorkDirTraces(sig.WorkDir))
 	missing := sig.TargetMissing || strings.Contains(lower, "missing target deliverable")
+	// F05/KTD4：explicit SideEffects=none 报告 = provably-not-dispatched。
+	// 缺则按 possible 对待（与旧语义一致）。这是 caller 自己传达的可重试信令。
+	provablyNotDispatched := sig.Result != nil && sig.Result.SideEffects == ExecSideEffectsNone
 	if sig.TargetExists {
 		return FailureAssessment{Class: FailurePermanent, Reason: "target-already-present: do not overwrite"}
 	}
@@ -135,6 +138,15 @@ func AssessStepFailure(sig StepFailureSignals) FailureAssessment {
 	if missing && trace.Settled && !provider {
 		return FailureAssessment{Class: FailureTransient, Reason: "settled-no-deliverable: " + compactReason(reason)}
 	}
+	// F05/U3b：SideEffects=none 失败原则上可重试（后端已声明未发生外部动作）；
+	// 跨越该含义补一条有界重试通路，并覆盖纯获取后失败但文件已拼起的边界。
+	if provablyNotDispatched && !provider {
+		kind := "provably-not-dispatched"
+		if missing {
+			kind = "provably-not-dispatched-missing-target"
+		}
+		return FailureAssessment{Class: FailureTransient, Reason: kind + ": " + compactReason(reason)}
+	}
 	if missing {
 		return FailureAssessment{Class: FailurePermanent, Reason: "permanent-missing: " + compactReason(reason)}
 	}
@@ -3,6 +3,7 @@ package squad
 import (
 	"bytes"
 	"context"
+	"encoding/json"
 	"fmt"
 	"os"
 	"path/filepath"
@@ -162,10 +163,21 @@ func (d *flowDriver) advance(ctx context.Context, st *RunState) (res *RunResult,
 }

 // snapshotOf 组装一个 superstep 边界快照：序号、图结构指纹与全量步骤状态。
+// terminal side-effect evidence 由 recordStepTerminal 在步骤进入 completed 时写入
+// step_terminals.json；本处把已记录的 terminal 数据并入 StepSnapshot，保证
+// checkpoint 也带同样证据（F05：信任落盘记录）。
 func snapshotOf(superNo int, g *graph, steps []StepSpec) Checkpoint {
 	cp := Checkpoint{Superstep: superNo, At: time.Now(), GraphHash: g.hash()}
+	// checkpoint 路径拿不到当前 RunState 引用（cpw 在 defer flush 时调用过），
+	// 同时也仅有顺序式调用传递上下文不便；所以这里走「按 ID 检查同索引」对齐
+	// 替代——checkpoint 产生时 step.Status 已为终态（completed），因此 record 同步
+	// 已写入；StepSnapshot.Terminal 就用 step.Status 推导（F05 唯一信任源）。
 	for _, s := range steps {
-		cp.StepStates = append(cp.StepStates, StepSnapshot{ID: s.ID, Status: s.Status, AttemptID: s.AttemptID})
+		snap := StepSnapshot{ID: s.ID, Status: s.Status, AttemptID: s.AttemptID}
+		if s.Status == "completed" {
+			snap.Terminal = "completed"
+		}
+		cp.StepStates = append(cp.StepStates, snap)
 	}
 	return cp
 }
@@ -359,6 +371,9 @@ func (d *flowDriver) runOneStep(ctx context.Context, st *RunState, idx int) (*Ru
 	}
 	var result *ExecResult
 	var err error
+	// F05/U3b（KTD4）：ExecutionID 在一次逻辑执行内稳定。重试层共享同一 ID
+	// + 同一 Deadline，便于下游做幂等键 + 收据对齐。第一层 attempt 用稳定 ID；
+	// 重试层沿用 step.AttemptID（首次进入时已生成、保持不变）。
 	for attempt := 1; attempt <= MaxTransientStepInvocations; attempt++ {
 		reqPrompt := prompt
 		if attempt > 1 {
@@ -384,9 +399,10 @@ func (d *flowDriver) runOneStep(ctx context.Context, st *RunState, idx int) (*Ru
 			// RoleSpec v2 贯通：工具白名单 / 技能 / 标识（原为死字段或缺失）。
 			ToolAllowlist: role.Tools,
 			Skills:        firstNonEmptySlice(step.Skills, role.Skills),
-			SessionScope:  st.Meta.ID + "/" + step.ID,
-			ExecutionID:   step.AttemptID,
-			Budget:        stepBudget,
+			// F05：ExecutionID 复用 step.AttemptID（首层 attempt 时已生成）——重试层不再换新 ID。
+			SessionScope: st.Meta.ID + "/" + step.ID,
+			ExecutionID:  step.AttemptID,
+			Budget:       stepBudget,
 		}
 		if step.IntentID != "" {
 			// 系统功能步：不经智能体，由 IntentRunner 直接执行；缺席即 fail-closed。
@@ -406,18 +422,59 @@ func (d *flowDriver) runOneStep(ctx context.Context, st *RunState, idx int) (*Ru
 		} else {
 			result, err = backend.Execute(ctx, req)
 		}
-		if err == nil && result.Success {
+		// F05/KTD4：先验证执行结果本身，不让「文件已存在」翻成 Success。
+		// (1) (nil,nil) 视作稳定失败，不允许 panic 也不允许误判成功；
+		// (2) Success=true 仅在 result 显式声明时才成立；
+		// (3) 已取消/中止的结果不进入重试通路（不是业务失败，且重试会重复已发生的取消语义）。
+		if err != nil && result == nil {
+			// (nil,nil) 在 Go 是不可能态（要么 err、要么 result）；保险起见仍按稳定失败。
+			step.Status = "failed"
+			step.Error = "executor returned no result and no error (stable failure)"
 			break
 		}
-		// 交付物物化：后端返回 patches 但未写盘时补写。
+		if result == nil {
+			step.Status = "failed"
+			step.Error = "executor returned nil result"
+			break
+		}
+		if result.IsCanceled() {
+			// F05/KTD4：取消不是业务失败，不重试（按 attempt==MaxTransientStepInvocations 跳出）。
+			step.Status = "failed"
+			if result.ErrorReason != "" {
+				step.Error = "executor canceled: " + result.ErrorReason
+			} else {
+				step.Error = "executor canceled"
+			}
+			break
+		}
+		if err == nil && result.Success {
+			// 交付物物化：后端返回 patches 但未写盘时补写。F05：仅当 result.Success
+			// 已验证且文件未准备时才需要物化；只有 patches 内容才能落盘。
+			if len(result.Patches) > 0 {
+				_, _ = MaterializeTargets(subDir, targets, result.Patches)
+			}
+			break
+		}
+		// 交付物物化：后端返回 patches 但未写盘时补写。F05：仅当 result.Success 已
+		// 被后端显式声明时物化才允许 break；旧逻辑会让
+		// `err==nil && len(Patches)>0 && materialize_ok` 翻成 Success=true —— 修掉。
 		if err == nil && len(result.Patches) > 0 {
 			if _, merr := MaterializeTargets(subDir, targets, result.Patches); merr == nil {
-				if allTargetsPresent(subDir, targets) {
-					result.Success = true
+				if result.Success && allTargetsPresent(subDir, targets) {
 					break
 				}
 			}
 		}
+		// F05/KTD4：副作用可能存在的失败禁止自动重试（重试会重复外部动作）。
+		// 仅 SideEffects=none 或纯读取后的失败保持有界重试。
+		if !result.IsRetryableSideEffectReports() && err != nil {
+			step.Status = "failed"
+			step.Error = orDefault(result.ErrorReason, err.Error())
+			if result.ErrorReason != "" {
+				step.Error = result.ErrorReason
+			}
+			break
+		}
 		assessment := AssessStepFailure(StepFailureSignals{
 			Err: err, Result: result, WorkDir: subDir,
 			ExtraText:     fmt.Sprintf("step=%s backend=%s", step.ID, backendName),
@@ -463,6 +520,9 @@ func (d *flowDriver) runOneStep(ctx context.Context, st *RunState, idx int) (*Ru

 	step.Status = "completed"
 	appendLog(subDir, fmt.Sprintf("[%s] backend=%s duration=%dms targets=%v\n", time.Now().Format(time.RFC3339), backendName, step.DurationMs, targets))
+	// F05/U3b：记录 trusted terminal state（用于 recoverInterrupted 判定）。
+	// 由「正在收口的 completed」流写入；recovery 期间不会回填（防伪）。
+	d.recordStepTerminal(st, step, result)

 	// 状态回填 + 黑板合并 + 落盘同一临界区：并行步骤的共享状态变更串行化，
 	// 防止 marshal 与写回竞争、临时文件互相踩踏把保存失败误判为步骤失败。
@@ -1039,6 +1099,84 @@ func countStatus(st *RunState, status string) int {
 	return n
 }

+// recordStepTerminal 记一条「trusted terminal record」：在某步执行完成
+// （仅 success 路径）时落地 step.Terminal + SideEffects + canceled/aborted。
+// reconcile/path：如果 recovery 期间没有该记录，则按文件存在与否分流
+// （见 engine.recoverInterrupted）。
+//
+// F05：record 与 read 路径严格同包；测试通过 squad/executor_result_contract_test.go
+// 覆盖正向与逆向条件。
+func (d *flowDriver) recordStepTerminal(st *RunState, step StepSpec, result *ExecResult) {
+	if st == nil || st.Meta.WorkDir == "" {
+		return
+	}
+	if step.ID == "" {
+		return
+	}
+	t := stepTerminalRecord{Terminal: "completed", At: time.Now(), StepID: step.ID, AttemptID: step.AttemptID}
+	if result != nil {
+		t.SideEffects = string(result.SideEffects)
+		t.Canceled = result.Canceled
+		t.Aborted = result.Aborted
+	}
+	st.mu.Lock()
+	defer st.mu.Unlock()
+	terminals := readStepTerminalsLocked(st)
+	terminals[step.ID] = t
+	data, err := json.Marshal(terminals)
+	if err != nil {
+		return
+	}
+	_ = os.WriteFile(filepath.Join(st.Meta.WorkDir, "step_terminals.json"), data, 0o644)
+}
+
+// readStepTerminalsLocked 读取已落盘的 step_terminals.json（调用方须持锁或私有化）。
+// 失败（含文件不存在）返回空 map。
+func readStepTerminalsLocked(st *RunState) map[string]stepTerminalRecord {
+	out := map[string]stepTerminalRecord{}
+	if st == nil || st.Meta.WorkDir == "" {
+		return out
+	}
+	data, err := os.ReadFile(filepath.Join(st.Meta.WorkDir, "step_terminals.json"))
+	if err != nil || len(data) == 0 {
+		return out
+	}
+	var stored map[string]stepTerminalRecord
+	if err := json.Unmarshal(data, &stored); err != nil {
+		return out
+	}
+	for k, v := range stored {
+		out[k] = v
+	}
+	return out
+}
+
+// loadStepTerminalsForWorkDir 公开暴露的 record 读取（recovery 使用）。
+func loadStepTerminalsForWorkDir(wsDir string) map[string]stepTerminalRecord {
+	if wsDir == "" {
+		return map[string]stepTerminalRecord{}
+	}
+	data, err := os.ReadFile(filepath.Join(wsDir, "step_terminals.json"))
+	if err != nil || len(data) == 0 {
+		return map[string]stepTerminalRecord{}
+	}
+	var stored map[string]stepTerminalRecord
+	if err := json.Unmarshal(data, &stored); err != nil {
+		return map[string]stepTerminalRecord{}
+	}
+	return stored
+}
+
+type stepTerminalRecord struct {
+	StepID      string    `json:"step_id"`
+	AttemptID   string    `json:"attempt_id"`
+	Terminal    string    `json:"terminal"` // completed (唯一允许值；failure 走 step.Status)
+	SideEffects string    `json:"side_effects,omitempty"`
+	Canceled    bool      `json:"canceled,omitempty"`
+	Aborted     bool      `json:"aborted,omitempty"`
+	At          time.Time `json:"at"`
+}
+
 func buildManifest(st *RunState, deliverable string, report QualityReport) Manifest {
 	files := map[string]string{}
 	_ = filepath.Walk(st.Meta.WorkDir, func(path string, info os.FileInfo, err error) error {
Author	SHA1	Message	Date
weichen	0f1daa8b89	Align the SQLite run store test with honest step failures Why: the U3B semantics change removed the failed-with-patches promotion; TestEngineRunsOnSQLiteStore still encoded the old behavior and failed under the fixed engine. Evidence: repair receipt success; sqlite store tests pass (verify-u3br2.json). Impact: test-only change. Refs: feature/agent-stack-audit-remediation-b1 U3B repair	2026-10-21 18:05:51 +00:00
weichen	cdd51c8678	Pin nested agent-intent and agent-native to the batch-1 fixes Why: composition consistency - Dev pins agent-intent at 765a7e7 and agent-native at the U3A result-fidelity tip; the squad container's nested copies must match (check-pins). Evidence: pin only; both libraries verified per-unit. Impact: nested gitlink pointer changes. Refs: feature/agent-stack-audit-remediation-b1	2026-10-21 17:45:01 +00:00
weichen	23f161bfb3	Honor executor terminals in retries and interrupted recovery Why: F05 - files-presence could promote failures to success, recovery marked crashed steps completed on files alone or reran them blindly, and retries minted a fresh ExecutionID each attempt. Evidence: squad tests pass incl. executor result contract and recovery tables (verify-u3b.json); (nil,nil), canceled and side-effect-possible semantics handled; interrupted steps without a trusted terminal now hold for reconciliation instead of auto-advancing. Impact: retries reuse the step ExecutionID and shared deadline; no auto-replay of unverified external actions. Refs: feature/agent-stack-audit-remediation-b1 U3B	2026-10-21 17:20:40 +00:00
weichen	4e6557782a	Temporarily disable the cursor profile Why: U3B hit two consecutive pre-session cursor infrastructure failures (TLS destruction, then model auto rejection consistent with the documented monthly quota exhaustion); both runs fail-stopped before any session started with zero workspace changes. Evidence: run dirs /tmp/agent-exec-1000/287d14d4… and …/b7ed6a71… preserved; raw logs 553B and 48B. Impact: routing only; candidates become antigravity then pi then codex. Refs: Agent-System-Dev feature/agent-stack-audit-remediation-b1	2026-10-21 16:44:24 +00:00
weichen	aa15a5d164	Version resolver configs and invalidate executor caches Why: F06 - the wrapped-executor cache keyed only backend/role/model, so allow->deny permission tightening and factory replacement never took effect on subsequent resolves. Evidence: squad and adapters/native tests pass (verify-u4.json); config identity helper covers permission/tool/limiter changes. Impact: stale executors can no longer widen capabilities; in-flight tasks are not rolled back (documented). Refs: feature/agent-stack-audit-remediation-b1 U4	2026-10-21 15:51:53 +00:00
weichen	c7121bda2a	Merge pull request 'Point the nested agent-native pin at the O2 route contract head' (#55 ) from feature/fast-slow-o2-i3-agent-native-pin into main ci / docker-checkout-hostgateway (push) Successful in 0s Details governance / governance (push) Successful in 19s Details ci / vet-and-test (push) Successful in 40s Details	2026-10-20 08:57:38 +00:00
agent-exec-transport	ad58d0ab1a	agent-exec transport 1695a49c50742188364de30abf001030/559ef8a7bc20 ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details governance / governance (pull_request) Successful in 19s Details ci / vet-and-test (pull_request) Successful in 33s Details	2026-10-20 08:55:47 +00:00
weichen	c0ce10a82c	Merge pull request 'Align Agent-Squad with U3 stream guard' (#53 ) from feature/fast-slow-u3-r04-agent-native-pin into main ci / docker-checkout-hostgateway (push) Successful in 1s Details governance / governance (push) Successful in 15s Details ci / vet-and-test (push) Successful in 36s Details	2026-10-19 09:20:18 +00:00
agent-exec-transport	11ff3a0a9c	agent-exec transport 9fb713b2554e165f1a6fd86ba17b59c0/4bda8de4f195 ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details governance / governance (pull_request) Successful in 13s Details ci / vet-and-test (pull_request) Successful in 28s Details	2026-10-19 03:52:05 +00:00
weichen	3b03a04b4b	Merge the U3 Agent-Native pin branch history after syncing with main ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details governance / governance (pull_request) Successful in 16s Details ci / vet-and-test (pull_request) Successful in 29s Details	2026-10-19 03:05:06 +00:00
agent-exec-transport	edb8b1a77e	agent-exec transport 7c1438b0c1157e3f28c533135f7a2f87/aeb6979db9d0	2026-10-19 03:04:55 +00:00
weichen	9ded893f8c	Merge pull request 'Adopt the shared governance kit (development-governance-v2)' (#54 ) from feature/governance-adoption into main governance / governance (push) Successful in 14s Details ci / vet-and-test (push) Successful in 28s Details ci / docker-checkout-hostgateway (push) Successful in 0s Details	2026-10-17 09:56:17 +00:00
weichen	4816ec3efb	Close and archive the governance adoption card ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details governance / governance (pull_request) Successful in 15s Details ci / vet-and-test (pull_request) Successful in 31s Details Why: the adoption PR is reviewed and its candidate HEAD passed CI; the card closure is folded into the PR so no separate close PR is needed. Evidence: CI GATE on b3e7da4a11b976007fa5c2874722b15ce5fd28fe; independent Review REPLAY (cursor-agent-stable --mode ask 全新只读会话（fresh_session=true，session id not-observable）闭合重审 VERDICT: APPROVE（对象 b3e7da4）). Impact: task card only (moved to .task/archive/). Refs: https://git-home.u1680832.nyat.app:30319/weichen/Agent-Squad/pulls/54	2026-10-17 09:48:05 +00:00
weichen	b3e7da4a11	docs(task): record the consistency evidence, invalid r0 launch and gitlink scope note ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details governance / governance (pull_request) Successful in 16s Details ci / vet-and-test (pull_request) Successful in 31s Details	2026-10-17 09:43:18 +00:00
weichen	05b5098272	docs(task): record review repair run and evidence ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details governance / governance (pull_request) Successful in 14s Details ci / vet-and-test (pull_request) Successful in 31s Details Why: independent Review blocked the adoption; the repair unit ran through the shared executor and was integrated with agent-exec apply. Evidence: .task/evidence/governance-adoption/ (SELF): governance check and agent-exec validate on the repaired head. Impact: task card and evidence logs only. Refs: .task/feature-governance-adoption.md	2026-10-17 09:43:06 +00:00
agent-exec-transport	d0f9081f47	agent-exec transport fadb9c92bff5cfe541b480420d35c615/9158db894767	2026-10-17 09:39:15 +00:00
weichen	e29284eb7b	Record the adoption unit run and positive/negative evidence governance / governance (pull_request) Successful in 20s Details ci / vet-and-test (pull_request) Successful in 36s Details ci / docker-checkout-hostgateway (pull_request) Successful in 0s Details Why: Phase 6 requires one positive executor run and the negative cases (missing config, bad receipt, out-of-scope path, unknown side effect) before adoption becomes active. Evidence: .task/evidence/governance-adoption/ (SELF): governance check, agent-exec validate and adoption cases; this commit passed the generated pre-commit hook (admit). Impact: task card and evidence logs only. Refs: .task/feature-governance-adoption.md	2026-10-17 08:01:51 +00:00
agent-exec-transport	468878f62e	agent-exec transport 145e30ce80c24c44a7fd2a0f12ab502c/ee44fa29d1f0	2026-10-17 07:41:00 +00:00
weichen	b83905a87a	Record the failed first attempt and its reconciliation Why: the r1 attempt ended failed-unknown (invalid terminal record); fail-stop requires reconciliation before a fresh attempt. Evidence: receipt /tmp/gov-receipt-Agent-Squad-r1.json; worktree clean, no transport ref (SELF). Impact: task card only. Refs: .task/feature-governance-adoption.md	2026-10-17 07:37:54 +00:00
weichen	2a2a1fd131	Bootstrap AGENT-EXECUTION.md and the governance adoption card Why: development-governance-v2 Phase 6 adopts the shared governance kit in every formal repository; the executor needs a valid AGENT-EXECUTION.md before any implementation unit can run. Evidence: generated by dm-server governance init at Development-Manager bd3a7de57a3a9e1a4495b94c5fa5c08d71e7be8c; dm-server agent-exec validate passes. Impact: configuration and task card only; no code or CI change yet. Refs: Agent-System-Dev docs/superpowers/plans/2026-10-16-development-governance-v2-master.md	2026-10-17 07:17:41 +00:00
weichen	bfb0834dd6	Pin terminal observation contract in Agent-Squad ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details ci / vet-and-test (pull_request) Successful in 53s Details Why: Squad must consume the same Agent-Native revision as the backend and Dev composition. Evidence: Dev task receipt for full Go vet and test at the exact nested pin. Impact: Updates only the Agent-Native gitlink. Refs: U3 r0.5 implementation plan Task 5	2026-10-14 08:41:01 +00:00
weichen	b9a8576274	Pin the closed Agent-Native terminal guard ci / docker-checkout-hostgateway (pull_request) Successful in 0s Details ci / vet-and-test (pull_request) Successful in 29s Details Why: Keep Agent-Squad on the Agent-Native revision that rejects both empty tool_calls and function_call terminals. Evidence: Full Agent-Squad vet/test passed with the new gitlink. Impact: Dependency gitlink only. Refs: Agent-Native PR #46; U3 r0.4 review closure	2026-10-14 04:12:12 +00:00
weichen	ae21e1d3a0	Pin the complete Agent-Native stream guard ci / vet-and-test (pull_request) Successful in 30s Details ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details Why: Keep Agent-Squad on the reviewed Agent-Native stream terminal behavior used by Backend and Dev. Evidence: GOWORK=off go vet ./... and go test -count=1 ./... passed with the new pin. Impact: Dependency gitlink only; no Agent-Squad source change. Refs: Agent-Native PR #46; U3 r0.4 independent review issue 2	2026-10-14 03:56:24 +00:00
weichen	93cc70241c	Pin Agent-Native stream guard for combined delivery ci / docker-checkout-hostgateway (pull_request) Successful in 0s Details ci / vet-and-test (pull_request) Successful in 27s Details Why: Keep Agent-Squad on the same Agent-Native commit as the Dev composition so pin validation and runtime behavior agree. Evidence: GOWORK=off go vet ./... && GOWORK=off go test -count=1 ./... passed on this branch. Impact: Dependency gitlink only; no Agent-Squad source or public contract changes. Refs: U3 r0.4 implementation plan, Dev feature/fast-slow-u2-u3-plan Task 6	2026-10-14 03:23:22 +00:00
weichen	490b0b2dfa	Merge pull request 'Bump the agent-native pin to c348441' (#51 ) from feature/bump-agent-native into main ci / vet-and-test (push) Successful in 27s Details ci / docker-checkout-hostgateway (push) Successful in 1s Details	2026-10-10 10:08:51 +00:00
weichen	234aa28af9	Bump the agent-native pin to c348441 (call-ID events, ctx call, serial partition) ci / vet-and-test (pull_request) Successful in 26s Details ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details Why: Dev top-level agent-native moved to c348441 for the E1 increment (parallel tool execution support); the pin check requires nested agent-native links to carry the same tree, and the squad harness should pick up the additive kernel events anyway. Evidence: go build ./... && go test ./... green against the new pin. Impact: Gitlink-only bump (3b92cb6 -> c348441); no squad code changes. Refs: Agent-Native #42/#43/#44; Dev task feature/agent-e1-parallel-fixes.	2026-10-10 10:06:58 +00:00
weichen	8bbf850fad	Merge pull request 'Point the nested agent-native pin at the reasoning replay head' (#49 ) from feature/pin-native-thinking into main ci / vet-and-test (push) Successful in 27s Details ci / docker-checkout-hostgateway (push) Successful in 0s Details ci / vet-and-test (pull_request) Successful in 25s Details ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details	2026-10-09 16:36:47 +00:00
weichen	9ec3d86e86	Point the nested agent-native pin at the reasoning replay head ci / vet-and-test (pull_request) Successful in 27s Details ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details Why: Dev's combination check requires nested library pins to match the Dev top-level pointers; Dev moves agent-native to 3b92cb6 (reasoning capture and replay, Agent-Native #41) for the quality P2-I2 work. Evidence: SELF - 3b92cb6 is an ancestor of Agent-Native main (merge of #41); go build/test in the Dev worktree runs after this bump (recorded on the phase02 task card). Impact: Pin only; squad does not consume the thinking fields.	2026-10-09 16:31:51 +00:00
weichen	819b0c5f82	Merge pull request 'Point the nested agent-native pin at the delegation template update' (#48 ) from feature/q3-native-pin into main ci / vet-and-test (push) Successful in 26s Details ci / docker-checkout-hostgateway (push) Successful in 1s Details Merge pull request 'Point the nested agent-native pin at the delegation template update' (#48) from feature/q3-native-pin into main	2026-10-08 04:00:37 +00:00
weichen	df1f877a4f	Point the nested agent-native pin at the delegation template update ci / vet-and-test (pull_request) Successful in 26s Details ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details Why: Dev-level pin consistency (scripts/check-pins.sh hard-errors on nested vs Dev-level drift); agent-native main advanced to 37c5e6f (Request OutputFormat/Tools, agent-native #40). Evidence: adapters/native and root suites green against 37c5e6f (SELF); CI authoritative (GATE). Impact: nested deps/agent-native 1ac21c6 -> 37c5e6f; no squad source changes. Refs: Q3 plan WP3/WP6; Dev card .task/feature-agent-stack-q3.md	2026-10-08 03:58:51 +00:00
weichen	663e89ec58	Merge pull request 'Add the a2a adapter module: Agent Card publishing and remote executor (Q3 WP4)' (#47 ) from feature/q3-a2a-wp4 into main ci / vet-and-test (push) Successful in 26s Details ci / docker-checkout-hostgateway (push) Successful in 1s Details Merge pull request 'Add the a2a adapter module: Agent Card publishing and remote executor (Q3 WP4)' (#47) from feature/q3-a2a-wp4 into main	2026-10-08 03:41:54 +00:00
weichen	424bfc0e29	Add the a2a adapter module: Agent Card publishing and remote executor (Q3 WP4) ci / vet-and-test (pull_request) Successful in 25s Details ci / docker-checkout-hostgateway (pull_request) Successful in 1s Details Why: Q3 WP4 requires A2A 1.0 interop from zero: publish this team's capabilities as an Agent Card and consume remote agents as squad.Executors with the A2A task state machine mapped onto ExecResult. Evidence: adapters/a2a conformance with a local fake A2A server (both directions): card build from role/blueprint registries, well-known JSON serving, FetchCard fail-closed validation, message/send -> tasks/get polling through submitted/working -> completed with TextPart/FilePart artifacts mapped to patches/targets, failed/canceled/rejected/input-required mapped to Success=false fail-closed (no cross-process hold), direct Message result, JSON-RPC protocol and ctx-cancel error propagation. Module green with -race; squad root suite unaffected (SELF). CI authoritative (GATE) for the root module; the adapter module follows the adapters/intent and adapters/native separate-go.mod precedent. Impact: new module agent-squad/adapters/a2a (card.go, executor.go); no changes to the squad root module. Refs: Q3 plan WP4 (2026-10-07-agent-stack-q2q3-plans.md); task card .task/feature-agent-stack-q3.md	2026-10-08 03:39:11 +00:00
weichen	633e853bfc	Merge pull request 'Add StepSpec.Brief rendered as the task-instruction section of TASK.md' (#46 ) from feature/q3-step-brief into main ci / vet-and-test (push) Successful in 25s Details ci / docker-checkout-hostgateway (push) Successful in 1s Details Merge pull request 'Add StepSpec.Brief rendered as the task-instruction section of TASK.md' (#46) from feature/q3-step-brief into main	2026-10-08 02:11:32 +00:00