Compare commits

..
Author SHA1 Message Date
agent-exec-transport 4a343c68d4 agent-exec transport 595670b7422a089295bccb99bc83f217/c91935ab769c
ci / test (pull_request) Successful in 9m45s
ci / test-pg (pull_request) Successful in 13m13s
ci / svc-realcombo (pull_request) Failing after 3m35s
2026-10-23 16:06:19 +00:00
agent-exec-transport 7df19b3f58 agent-exec transport b36b5f5c4c6339d6aa4534cc2f5858db/50d4a643b8cc
ci / svc-realcombo (pull_request) Failing after 3m34s
ci / test (pull_request) Successful in 9m53s
ci / test-pg (pull_request) Successful in 14m3s
2026-10-23 15:57:44 +00:00
agent-exec-transport 1191cee681 agent-exec transport e187af3b4d0e2c603177b73dbbe380dc/826db5d9de5c
ci / test (pull_request) Successful in 12m4s
ci / svc-realcombo (pull_request) Failing after 5m53s
ci / test-pg (pull_request) Failing after 6m24s
2026-10-23 15:41:01 +00:00
agent-exec-transport 0df91e1ead agent-exec transport efb497ddb56c8a0766c1a263c0fe3fdc/566cc005aff7
ci / svc-realcombo (pull_request) Failing after 3m24s
ci / test (pull_request) Successful in 12m45s
ci / test-pg (pull_request) Successful in 14m49s
2026-10-23 14:53:32 +00:00
agent-exec-transport 7f79399354 agent-exec transport c9aa40df76eaa815b4d0a80342eb8343/7516af9617f3
ci / svc-realcombo (pull_request) Failing after 4m46s
ci / test (pull_request) Successful in 10m42s
ci / test-pg (pull_request) Successful in 14m14s
2026-10-23 14:43:52 +00:00
agent-exec-transport 90d3b64fed agent-exec transport 2112c47d870b208e6bc7bc12f74a3aa1/e0361a025a40
ci / svc-realcombo (pull_request) Failing after 7m23s
ci / test (pull_request) Successful in 10m55s
ci / test-pg (pull_request) Successful in 15m42s
2026-10-23 13:49:06 +00:00
agent-exec-transport 0951822361 agent-exec transport 827954082c92cad23c4268e65a022402/3c2abc731b47
ci / test (pull_request) Successful in 15m30s
ci / test-pg (pull_request) Successful in 20m34s
ci / svc-realcombo (pull_request) Failing after 6m28s
2026-10-23 13:23:23 +00:00
agent-exec-transport ecbed09074 agent-exec transport 2c39b240737178c8b0a6a300a7815f25/08e38972e379
ci / svc-realcombo (pull_request) Failing after 4m24s
ci / test (pull_request) Successful in 14m41s
ci / test-pg (pull_request) Successful in 20m7s
2026-10-23 13:02:59 +00:00
agent-exec-transport e66d6324eb agent-exec transport 2e4df9d51aa38cc1fef8f26a94ff6cc3/aa27595c71fd
ci / test (pull_request) Successful in 12m31s
ci / test-pg (pull_request) Successful in 18m18s
ci / svc-realcombo (pull_request) Failing after 3m7s
2026-10-23 12:26:35 +00:00
agent-exec-transport 26ad9a732f agent-exec transport 113b493c2e70a008ddf78512ddcd382c/0626f80ffdfe
ci / svc-realcombo (pull_request) Failing after 2m28s
ci / test (pull_request) Successful in 11m43s
ci / test-pg (pull_request) Successful in 15m56s
2026-10-23 11:51:49 +00:00
weichen 0650e81133 Merge pull request 'Recall mis-execution fixes: negation-aware matching and fail-closed slots' (#717) from feature/recall-misexec-fixes into main
ci / svc-realcombo (push) Successful in 3m36s
ci / test (push) Successful in 9m26s
ci / test-pg (push) Successful in 12m5s
2026-10-23 05:39:42 +00:00
weichen 9aa18eb30b Merge authoritative main into the task branch
ci / svc-realcombo (pull_request) Successful in 2m53s
ci / test (pull_request) Successful in 9m23s
ci / test-pg (pull_request) Successful in 12m50s
Why: include the current integration baseline before merge.

Evidence: fetched origin/main=a6485795.

Impact: integrate upstream changes without rewriting task history.

Refs: task-sync
2026-10-23 05:25:48 +00:00
weichen a648579519 Merge pull request 'Snapshot all RDM pins in the e2e build' (#718) from fix/e2e-pins-snapshot into main
ci / svc-realcombo (push) Successful in 3m38s
ci / test (push) Successful in 10m8s
ci / test-pg (push) Successful in 14m38s
2026-10-23 05:14:52 +00:00
weichen 226bf8d38c Merge authoritative main into the task branch
ci / svc-realcombo (pull_request) Successful in 3m24s
ci / test (pull_request) Successful in 10m43s
ci / test-pg (pull_request) Successful in 14m5s
Why: include the current integration baseline before merge.

Evidence: fetched origin/main=3dae91ec.

Impact: integrate upstream changes without rewriting task history.

Refs: task-sync
2026-10-23 05:10:50 +00:00
agent-exec-transport ec89964880 agent-exec transport 00b39ef7c68383bc27c5a700f1f69f10/4f97c052dc11
ci / svc-realcombo (pull_request) Successful in 3m2s
ci / test (pull_request) Successful in 10m1s
ci / test-pg (pull_request) Successful in 15m8s
2026-10-23 04:58:50 +00:00
weichen 27cd65e450 Fold-close the task card before merge
ci / svc-realcombo (pull_request) Successful in 2m56s
ci / test (pull_request) Successful in 9m48s
ci / test-pg (pull_request) Successful in 13m27s
Why: close the task card inside its own PR so landing it needs no post-merge backfill or separate close PR.
Evidence: .task/feature-recall-misexec-fixes.md -> .task/archive/feature-recall-misexec-fixes.md; status 已完成(合入前闭合); PR #717 https://git-home.u1680832.nyat.app:30319/weichen/Agent-System-Dev/pulls/717.
Impact: task metadata only; the card is archived within PR #717 and lands with its merge commit.
Refs: task.sh close --self
2026-10-23 04:56:44 +00:00
weichen b497f75e03 Record the mis-execution review approval in the task card
ci / test-pg (pull_request) Successful in 12m0s
ci / svc-realcombo (pull_request) Successful in 2m47s
ci / test (pull_request) Successful in 9m13s
Why: independent review approved the rule-layer fixes (0 blockers,
0 majors).
Evidence: review receipt /tmp/agent-stack-b2/reviewm-run.json.
Impact: task metadata + review report.
Refs: PR #717
2026-10-23 04:44:18 +00:00
agent-exec-transport 3b78ccb6b0 agent-exec transport d6c185868358ea9104955fd065e9b9c0/57e463862f86 2026-10-23 04:43:21 +00:00
weichen 8e3f9edb24 Record the mis-execution fix results in the task card
ci / test (pull_request) Successful in 11m7s
ci / test-pg (pull_request) Successful in 13m55s
ci / svc-realcombo (pull_request) Successful in 3m26s
Why: audit trail for the rule-layer fixes and the pin cascade.
Evidence: verify-misexec2.json (mis_exec=0, 37/37 pass, race) and
verify-misexec-combo2.json exit=0.
Impact: task metadata only.
Refs: feature/recall-misexec-fixes
2026-10-23 04:23:22 +00:00
weichen 56ca0ee7c7 Repin modules/backend to the corrected nested pins tip
Why: the backend branch tip carries the corrected nested
intent/native/squad pins for the mis-execution fix composition.
Evidence: integration transport (misexec-git3-run.json); full
combination re-run below.
Impact: one component gitlink.
Refs: feature/recall-misexec-fixes
2026-10-23 04:14:56 +00:00
weichen 3dae91ecf4 Merge pull request 'Upgrade the lifecycle RDM pin to increment D' (#715) from fix/rdm-pin-increment-d into main
ci / svc-realcombo (push) Successful in 2m46s
ci / test (push) Successful in 9m10s
ci / test-pg (push) Successful in 11m55s
2026-10-23 04:05:09 +00:00
weichen c45fbf9569 Advance native, intent, squad and backend pins to the mis-execution fix tips
Why: land the A3-baseline-driven rule-layer fixes (negation-aware
matching, specificity tie-break, fail-closed slots) across the
composition.
Evidence: unit transport (misexec-git-run.json); eval mis_exec=0 with
37/37 pass (verify-misexec2.json); full combination re-run below.
Impact: four component gitlinks.
Refs: feature/recall-misexec-fixes
2026-10-23 04:00:02 +00:00
weichen 29d97aec39 Record the mis-execution fix design in the task card
Why: fix the design (negation-aware matching in compiledSpec.evaluate,
specificity tie-break in adjudication, slot multi-value/charset
invalidity) with the eval corpus as acceptance.
Evidence: A3 baseline analysis (evaluation/BASELINE.md).
Impact: task metadata only.
Refs: feature/catalog-recall-eval-a3 (baseline)
2026-10-23 03:49:24 +00:00
weichen 06d0afb76c Record the task card for feature-recall-misexec-fixes
Why: establish the authorized task workspace and its initial baseline.

Evidence: preserved precheck receipt in .task/feature-recall-misexec-fixes.md; R0=bccb88f294038ce52a37fc5592a6daebd10e88b4.

Impact: task metadata only; the feature workspace starts at bccb88f294.

Refs: .task/feature-recall-misexec-fixes.md
2026-10-23 03:45:01 +00:00
8 changed files with 274 additions and 24 deletions
+49 -17
View File
@@ -35,7 +35,9 @@ jobs:
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")
# Allowed SSH prefixes: legacy nyat.app:31411 and migrated gitea.naitangbaobao.cn:52222.
SSH = "ssh://git@git-ssh.u1680832.nyat.app:31411/weichen/"
SSH_MIGRATED = "ssh://git@gitea.naitangbaobao.cn:52222/weichen/"
# Reviewed allowlist, independent of candidate .gitmodules. Add repositories
# here only with an explicit review of the checkout credential's scope.
REPOSITORIES = {
@@ -124,7 +126,7 @@ jobs:
if not repository or fields.get("path") != name:
raise ValueError("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}:
if fields.get("url") not in {SSH + repository + ".git", SSH_MIGRATED + repository + ".git", trusted_url}:
raise ValueError("Submodule URL is outside the reviewed allowlist")
for name in entries:
url = f"{SERVER}/weichen/{REPOSITORIES[name]}.git"
@@ -244,7 +246,9 @@ jobs:
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")
# Allowed SSH prefixes: legacy nyat.app:31411 and migrated gitea.naitangbaobao.cn:52222.
SSH = "ssh://git@git-ssh.u1680832.nyat.app:31411/weichen/"
SSH_MIGRATED = "ssh://git@gitea.naitangbaobao.cn:52222/weichen/"
# Reviewed allowlist, independent of candidate .gitmodules. Add repositories
# here only with an explicit review of the checkout credential's scope.
REPOSITORIES = {
@@ -333,7 +337,7 @@ jobs:
if not repository or fields.get("path") != name:
raise ValueError("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}:
if fields.get("url") not in {SSH + repository + ".git", SSH_MIGRATED + repository + ".git", trusted_url}:
raise ValueError("Submodule URL is outside the reviewed allowlist")
for name in entries:
url = f"{SERVER}/weichen/{REPOSITORIES[name]}.git"
@@ -425,19 +429,22 @@ jobs:
set -euo pipefail
# 凭据纪律与 Checkout 块同源:经 GIT_CONFIG_* 环境注入 URL 级 token 头,
# 不落文件/命令行;allowlist 扩项 = Agent-System-RDM 及其 pins(同一
# 组件集合),由本 PR 评审背书。SSH pins URL 以 insteadOf 改写到受信
# https 基址后再取。只追加 pin 的三个 40 位 hex,不 source pin 文件。
# 组件集合),由本 PR 评审背书。SSH pins URL(新旧两形态均改写)以
# insteadOf 改写到受信 https 基址后再取。只追加 pin 的三个 40 位 hex,
# 不 source pin 文件。
SERVER="$(printf '%s' "$GITHUB_SERVER_URL" | sed 's:/*$::')"
scheme="${GITHUB_SERVER_URL%%://*}"
export GIT_CONFIG_COUNT=4
export GIT_CONFIG_COUNT=5
export GIT_CONFIG_KEY_0="protocol.$scheme.allow"
export GIT_CONFIG_VALUE_0="always"
export GIT_CONFIG_KEY_1="http.$SERVER/.extraheader"
export GIT_CONFIG_VALUE_1="AUTHORIZATION: token $SUBMODULE_TOKEN"
export GIT_CONFIG_KEY_2="url.$SERVER/weichen/.insteadOf"
export GIT_CONFIG_VALUE_2="ssh://git@git-ssh.u1680832.nyat.app:31411/weichen/"
export GIT_CONFIG_KEY_3="core.hooksPath"
export GIT_CONFIG_VALUE_3="/dev/null"
export GIT_CONFIG_KEY_3="url.$SERVER/weichen/.insteadOf"
export GIT_CONFIG_VALUE_3="ssh://git@gitea.naitangbaobao.cn:52222/weichen/"
export GIT_CONFIG_KEY_4="core.hooksPath"
export GIT_CONFIG_VALUE_4="/dev/null"
git clone --depth=1 "$SERVER/weichen/Agent-System-RDM.git" rdm
mapfile -t pins < <(python3 - <<'PY'
import pathlib
@@ -476,6 +483,7 @@ jobs:
-u GIT_CONFIG_KEY_1 -u GIT_CONFIG_VALUE_1 \
-u GIT_CONFIG_KEY_2 -u GIT_CONFIG_VALUE_2 \
-u GIT_CONFIG_KEY_3 -u GIT_CONFIG_VALUE_3 \
-u GIT_CONFIG_KEY_4 -u GIT_CONFIG_VALUE_4 \
-u SUBMODULE_TOKEN \
"$@"
}
@@ -504,7 +512,8 @@ jobs:
done
unset GIT_CONFIG_COUNT GIT_CONFIG_KEY_0 GIT_CONFIG_VALUE_0 \
GIT_CONFIG_KEY_1 GIT_CONFIG_VALUE_1 GIT_CONFIG_KEY_2 \
GIT_CONFIG_VALUE_2 GIT_CONFIG_KEY_3 GIT_CONFIG_VALUE_3
GIT_CONFIG_VALUE_2 GIT_CONFIG_KEY_3 GIT_CONFIG_VALUE_3 \
GIT_CONFIG_KEY_4 GIT_CONFIG_VALUE_4
unset SUBMODULE_TOKEN
- name: Install Node 22
run: |
@@ -569,7 +578,9 @@ jobs:
parsed_server = urlsplit(SERVER)
if parsed_server.scheme not in {"http", "https"} or not parsed_server.netloc:
sys.exit("Checkout server must be the platform-injected http(s) URL")
# Allowed SSH prefixes: legacy nyat.app:31411 and migrated gitea.naitangbaobao.cn:52222.
SSH = "ssh://git@git-ssh.u1680832.nyat.app:31411/weichen/"
SSH_MIGRATED = "ssh://git@gitea.naitangbaobao.cn:52222/weichen/"
# Reviewed allowlist, independent of candidate .gitmodules. Add repositories
# here only with an explicit review of the checkout credential's scope.
REPOSITORIES = {
@@ -658,7 +669,7 @@ jobs:
if not repository or fields.get("path") != name:
raise ValueError("Submodule path is outside the reviewed allowlist")
trusted_url = f"{SERVER}/weichen/{repository}.git"
if fields.get("url") not in {SSH + repository + ".git", trusted_url}:
if fields.get("url") not in {SSH + repository + ".git", SSH_MIGRATED + repository + ".git", trusted_url}:
raise ValueError("Submodule URL is outside the reviewed allowlist")
for name in entries:
url = f"{SERVER}/weichen/{REPOSITORIES[name]}.git"
@@ -713,19 +724,21 @@ jobs:
set -euo pipefail
# 凭据纪律与 Checkout 块同源:经 GIT_CONFIG_* 环境注入 URL 级 token 头,
# 不落文件/命令行;allowlist 扩项 = Agent-System-RDM 及其 pins(同一
# 组件集合),由本 PR 评审背书。SSH pins URL 以 insteadOf 改写到受信
# https 基址后再取。
# 组件集合),由本 PR 评审背书。SSH pins URL(新旧两形态均改写)以
# insteadOf 改写到受信 https 基址后再取。
SERVER="$(printf '%s' "$GITHUB_SERVER_URL" | sed 's:/*$::')"
scheme="${GITHUB_SERVER_URL%%://*}"
export GIT_CONFIG_COUNT=4
export GIT_CONFIG_COUNT=5
export GIT_CONFIG_KEY_0="protocol.$scheme.allow"
export GIT_CONFIG_VALUE_0="always"
export GIT_CONFIG_KEY_1="http.$SERVER/.extraheader"
export GIT_CONFIG_VALUE_1="AUTHORIZATION: token $SUBMODULE_TOKEN"
export GIT_CONFIG_KEY_2="url.$SERVER/weichen/.insteadOf"
export GIT_CONFIG_VALUE_2="ssh://git@git-ssh.u1680832.nyat.app:31411/weichen/"
export GIT_CONFIG_KEY_3="core.hooksPath"
export GIT_CONFIG_VALUE_3="/dev/null"
export GIT_CONFIG_KEY_3="url.$SERVER/weichen/.insteadOf"
export GIT_CONFIG_VALUE_3="ssh://git@gitea.naitangbaobao.cn:52222/weichen/"
export GIT_CONFIG_KEY_4="core.hooksPath"
export GIT_CONFIG_VALUE_4="/dev/null"
git clone --depth=1 "$SERVER/weichen/Agent-System-RDM.git" rdm
# 凭据环境须保持到 pins 子模块取完(insteadOf + token 头覆盖嵌套 SSH URL),
# 之后再 unset。
@@ -733,20 +746,39 @@ jobs:
echo "RDM commit: $(git -C rdm rev-parse HEAD)"
unset GIT_CONFIG_COUNT GIT_CONFIG_KEY_0 GIT_CONFIG_VALUE_0 \
GIT_CONFIG_KEY_1 GIT_CONFIG_VALUE_1 GIT_CONFIG_KEY_2 \
GIT_CONFIG_VALUE_2 GIT_CONFIG_KEY_3 GIT_CONFIG_VALUE_3
GIT_CONFIG_VALUE_2 GIT_CONFIG_KEY_3 GIT_CONFIG_VALUE_3 \
GIT_CONFIG_KEY_4 GIT_CONFIG_VALUE_4
unset SUBMODULE_TOKEN
make -C rdm build
- name: Run real-combination lifecycle gates (SQLite, U3, required)
env:
SUBMODULE_TOKEN: ${{ secrets.SUBMODULE_TOKEN }}
GIT_TRACE: "1"
SVC_RDM_ORIGIN: ${{ github.server_url }}/weichen/Agent-System-RDM.git
run: |
set -euo pipefail
# 路径变量在脚本内解析(YAML env: 不做 shell 展开)。
export SVC_TEST_REAL=1
env -u SUBMODULE_TOKEN make -C rdm product
export SVC_REAL_BIN="$PWD/bin/agent-system"
export SVC_REAL_RDM_BIN="$PWD/rdm/bin/agent-system-rdm"
export SVC_RDM_ORIGIN="$PWD/rdm"
export SVC_REAL_RDM_BIN="$PWD/rdm/bin/agent-system-rdm-product"
export SVC_RDM_SRC_DIR="$PWD/rdm"
# CI 无 SSH key:把受信基址与 URL 级 token 头经环境交给套件(仅环境传递)。
export SVC_CI_GIT_SERVER="${GITHUB_SERVER_URL%/}"
export SVC_CI_GIT_TOKEN="${SUBMODULE_TOKEN}"
# 套件经 SVC_RDM_SRC_DIR 物化嵌套 pins 时,RDM .gitmodules 的 SSH URL
# (新旧两形态)无凭据;按本 job Clone 步骤同源 insteadOf 改写到受信基址并带 token 头。
SERVER="$(printf '%s' "$GITHUB_SERVER_URL" | sed 's:/*$::')"
scheme="${GITHUB_SERVER_URL%%://*}"
export GIT_CONFIG_COUNT=5
export GIT_CONFIG_KEY_0="protocol.$scheme.allow"
export GIT_CONFIG_VALUE_0="always"
export GIT_CONFIG_KEY_1="http.$SERVER/.extraheader"
export GIT_CONFIG_VALUE_1="AUTHORIZATION: token $SUBMODULE_TOKEN"
export GIT_CONFIG_KEY_2="url.$SERVER/weichen/.insteadOf"
export GIT_CONFIG_VALUE_2="ssh://git@git-ssh.u1680832.nyat.app:31411/weichen/"
export GIT_CONFIG_KEY_3="url.$SERVER/weichen/.insteadOf"
export GIT_CONFIG_VALUE_3="ssh://git@gitea.naitangbaobao.cn:52222/weichen/"
export GIT_CONFIG_KEY_4="core.hooksPath"
export GIT_CONFIG_VALUE_4="/dev/null"
python3 -m unittest tests.test_svc_realcombo -v
@@ -0,0 +1,168 @@
# feature/recall-misexec-fixes 任务卡
任务主记录:PR #717 https://git-home.u1680832.nyat.app:30319/weichen/Agent-System-Dev/pulls/717;状态:已完成(合入前闭合);owner:weichen
## 基线与授权
- 权威仓:ssh://git@git-ssh.u1680832.nyat.app:31411/weichen/Agent-System-Dev.git;目标分支:main;R0(initial_base_sha)= bccb88f294038ce52a37fc5592a6daebd10e88b4
- 工作区:/home/weichen/projects/Agent-System-Dev-wt-feature-recall-misexec-fixes(worktree);分支:feature/recall-misexec-fixes
- 分支类型:standalone;root_target:main;immediate_target:main;accepting_requirements:false
- 需求分流:verdict=NEW;selected_initiative=无;requirement_sha256=5510782fe6debd449a832c3353d191c6efb76f50c77932c5adf387a2f0a27bcd;triage_receipt=/tmp/triage-misexec-1792727028.json;receipt_sha256=914545afa88ab6ce4cbe9bd48240e8a8b2f829c970d245c24fb1a7c0d8ca495c
- 授权与交付终点:用户已批准启动 A3 基线驱动的规则层修复(评测集暴露的 6 次误执行:语义否定被当作肯定 ×5、多意图平分误选寒暄 ×1;另有错误槽位未澄清 3 例)。范围内连续推进至合入;检索/调模型决策留待修复后基线。未授权:A4、部署。
- 设计决定(2026-10-22,host):
1. 否定修复放 compiledSpec.evaluate 匹配层:正命中若被前置否定词(不要/别/不用/先不要/先别/无需/不需要,紧邻或间隔≤2字)限定,且该 spec 的全部正命中均被否定 → reject "negated_match"(确定性,不逐 spec 配置)。
2. 平分裁决加确定性 tie-break:优先级相同 → 已抽取到槽位值多者优先(带槽位信号的动作胜过裸寒暄)→ id 字典序。不引入新配置。
3. 槽位抽取修正:抽取跨度内含多值分隔(和/还是/、/,)→ 视为多候选无效(clarify);规范化后仅剩符号/空 → 无效;字符集违规(如声明字母数字却抽出 abc_def 的部分截取)→ 无效。无效一律走 clarify,不 start。
4. 验收即评测集:上述用例从 record_only FAIL 翻为 PASS;overall mis_exec 须为 0;gen_agree 与其余类别不回退。
- 交付模式:goal-first(目标、范围和交付终点一次确认;范围内连续推进)
- 实施执行:agent-per-unit(配置:AGENT-EXECUTION.md)
- 合入执行:agent(已授权交付范围内,Review 通过、冲突解决且当前 HEAD 的 CI 通过后合入)
- 人工决策点:仅限目标/范围/验收变更、重大取舍、高后果操作或无法自动判定的业务验收
- 当前阶段:需求拆解;当前增量:I1(待定义)
- 流程依据:仓库根目录 policy/delivery.md(归档后仍按仓库根目录定位)
## 目标与非目标
**目标**:消除评测集 6 次误执行 + 3 例错误槽位未澄清:匹配层否定感知、裁决 specificity tie-break、槽位多值/字符集无效化;评测指标 mis_exec 归零且其余类别不回退;既有测试全绿。
**非目标**:检索/召回增强;模型/提示词调整;「刚才那个」跨会话指代解析(A1 会话域);真实语料;gen_agree 的 3 例生成器分歧(回放器行为,另行处理)。
## Agent 实施收据
路由与 fail-stop 规则见根目录 AGENT-EXECUTION.md 与 policy/delivery.md;每次尝试一行,原始日志留在源码外收据路径。
| 单元 ID | 预期范围 | 配置摘要 | 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 | NOT_RUN | — |
## 关联方向与规划
- 适用状态:待核对(无适用规划时说明原因)
- 主记录路径/版本:待核对(路径按仓库根目录定位,归档后不变;只引用唯一正文)
- 对应里程碑:待核对
## 需求分析
| ID | 描述 | 项目符合度及依据 | 优先级 | 验收条件 | 依赖 | 结论 |
|---|---|---|---|---|---|---|
| R1 | 待填写 | 待评估 | 待定 | 待定义 | 待确认 | 待评审 |
## 方案与计划
- 方案引用:待填写(含对象版本)
- 实施计划引用:待填写(含增量与需求 ID)
- 测试计划引用:待填写(含需求 ID 与验收条件)
## 阶段记录
| 阶段 | 对象版本 | 状态 | 决策/执行者 | 时间 | 依据 |
|---|---|---|---|---|---|
| 需求拆解 | 待定 | 待执行 | — | — | — |
| 需求评估 | 待定 | 待执行 | — | — | — |
| 需求表确认 | 待定 | 待执行 | — | — | — |
| 方案设计 | 待定 | 待执行 | — | — | — |
| 实施计划 | 待定 | 待执行 | — | — | — |
| 实施与自检 | 待定 | 待执行 | — | — | — |
| 正式测试与验收 | 待定 | 待执行 | — | — | — |
| 代码 Review | 待定 | 待执行 | — | — | — |
| 合入 | 待定 | 待执行 | — | — | — |
| 合入核验与收口 | 待定 | 待执行 | — | — | — |
## 探索试验与迭代
| 增量/试验 | 假设 | 范围 | 判断和停止条件 | 授权/决策依据 | 结果 | 变更影响 | 需重审补测 |
|---|---|---|---|---|---|---|---|
| 待定义 | — | — | — | 待核对 | NOT_RUN | 待评估 | 待评估 |
## 同步记录
| 节点 | sync_base_sha | tested_head_sha | checked_at |
|---|---|---|---|
| 启动 | bccb88f294038ce52a37fc5592a6daebd10e88b4 | bccb88f294038ce52a37fc5592a6daebd10e88b4 | 2026-10-23T03:45:01Z |
## 证据(Evidence)
<!-- CMD="<命令>" make evidence 写入;手抄/转写输出视为无效证据(AGENTS.md 证据与验证门禁)。 -->
| 时间(UTC) | 命令 | 被测HEAD | exit | 来源 | sha256(16) |
|---|---|---|---|---|---|
## 交付调度
- 调度状态:未入队;调度记录:无
- entry_id:无;attempt_id:无;窗口执行者:无
- 固定候选:无(base/head、组件与相关配置引用实际尝试)
- 调度依据:policy/delivery.md#合入调度
- 固定后证据:源码外 RECEIPT;实际结果后按授权原样导入和闭合
## 结果与验证
- 结果:
- 验证:<结论摘要 PASS/FAIL/SKIP/NOT_RUN 逐项;逐条证据由 make evidence 写入上方证据表>
- 影响与剩余项:
## 规划进展与下一步
- 本次推进/剩余:待核对(依据实际成果更新适用方案/规划)
- 首选下一任务:待核对(无适用规划、已完成或受阻时如实说明,不虚构任务)
- 推荐依据:待填写
- 前置条件/待确认:待核对
- 预期产物/验收条件:待定义
- 重复任务核对:待核对(已有/在途任务、PR 与规划记录)
- 启动授权:未授权(建议不自动启动新任务,仍须明确授权)
## 合入与清理
- 实施收据:misexec-rules-fix(cursor attempt1,否定感知匹配+specificity tie-break+槽位 fail-closed,intent 7f3ff6a,eval mis_exec=0/37 全过,verify-misexec2 exit=0);pin 级联嵌套提升(host 提交 N4/S4/B4')与 Dev 集成(misexec-git3-run.json)。
- 组合验证:全量 make test PASS(verify-misexec-combo2.json exit=0)。
- Review 记录:独立 CLI 评审单元 misexec-independent-review(收据 /tmp/agent-stack-b2/reviewm-run.json);结论 APPROVE;报告 .task/reviews/2026-10-22-misexec-fixes-independent-review.md;被审对象 8e3f9edb 组合 vs main
- 冲突记录:NOT_RUN(最新目标同步、冲突处置、组合兼容核验及复审/补测)
- CI 记录:NOT_RUN(当前 HEAD 完整 SHA、全部必需检查、运行链接、执行身份及 GATE 结果)
- 合入记录:NOT_RUN(交付授权范围、PR、被核验 HEAD、合入 SHA、目标分支及内容核验)
- 清理记录:NOT_RUN(授权依据、写者退出确认、worktree/分支/任务卡处置)
<!-- Local receipt: SELF. Digests detect corruption; they do not authenticate a reviewer or CI. -->
```json
{
"summary_schema": 1,
"source": "SELF",
"timestamp": "2026-10-23T03:44:02.622816+00:00",
"command": "make precheck",
"exit_code": 0,
"cwd": "/home/weichen/projects/Agent-System-Dev",
"before_head": "bccb88f294038ce52a37fc5592a6daebd10e88b4",
"after_head": "bccb88f294038ce52a37fc5592a6daebd10e88b4",
"branch": "main",
"target_branch": "main",
"target_head": "bccb88f294038ce52a37fc5592a6daebd10e88b4",
"root_target": "main",
"immediate_target": "main",
"before_tracked_diff_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
"after_tracked_diff_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
"output_tail": [
"make[1]: Entering directory '/home/weichen/projects/Agent-System-Dev'",
"[repo] branch=main head=bccb88f294",
"[remote] origin/main=bccb88f294 ahead=0 behind=0",
"[tree] clean submodules_drift=0 uninit=0",
"[worktrees] linked=7",
"[recycle] merged un-recycled branches: feature/exec-recovery-diagnostics-a2 fix/gitea-endpoint-facts",
"[initiatives] local cards (not authoritative): none",
"[hooks] core.hooksPath=scripts/hooks(pre-commit/post-merge 实际接线)",
"[facts] policy/facts.md 6 条事实均在有效期内",
"[required-reading] core/baseline.md, workflows/delivery.md §1, policy/delivery.md",
"[R0] bccb88f294(完整:bccb88f294038ce52a37fc5592a6daebd10e88b4)",
"[worktree] before-sync: recursive component checks passed",
"[verdict] READY — 保留本次 precheck receipt 后建任务工作区:PRECHECK_RECEIPT=/tmp/task-precheck.json make task-start NAME=feature/your-branch",
"make[1]: Leaving directory '/home/weichen/projects/Agent-System-Dev'"
],
"output_sha256": "6a6f70969d0b85288dde21bf2011b0485af1251a29fbb677ba7d7d9e94c79785",
"receipt_sha256": "b5fc6e81a92e8ecb97bbf51630a10608507425dfe27220efb75341b4f607bffd",
"receipt_path": "/home/weichen/.cache/agent-system-dev/evidence/b5fc6e81a92e8ecb97bbf51630a10608507425dfe27220efb75341b4f607bffd.json"
}
```
## 闭合记录(合入前折叠)
- 本卡在 PR #717(https://git-home.u1680832.nyat.app:30319/weichen/Agent-System-Dev/pulls/717)合入前闭合并归档,随该 PR 一同合入。
- 该 PR 的合并提交即本任务的落地记录。
- 合入后无需回填,也无需单独的闭合 PR。
@@ -0,0 +1,47 @@
# 独立代码 Review:feature/recall-misexec-fixes(mis-exec 规则层修复)
- 日期:2026-10-22(评审执行于 2026-10-23 UTC 的分支对象上)
- 评审人:独立 CLI 会话(未参与实施;本文件为本次唯一变更)
- 评审对象:`git diff bccb88f2...HEAD`(= origin/main...HEAD;本 worktree 无 origin remote,bccb88f2 为 main 顶端合并提交,是 feature 分支的 merge-base)
- 组件范围:deps/agent-intent @6aa4925..7f3ff6a(单提交 "Make intent matching negation-aware and slots fail closed")+ gitlink 级联(native d1f72aa→be556f4、squad a5afe9c8→95b387d、backend 8dc91c17→8031855e→bd5c0f3)+ 任务卡 `.task/feature-recall-misexec-fixes.md`
- 方法:全文精读 9 个变更文件 + 逐条手工追踪语料用例 + 对否定判定函数做 stdlib-only 逐字提取探针验证(/tmp,脱离工作区);未修改任何代码。
## 验证(本工作区内的执行情况)
- `git status --porcelain`:评审开始时为空(clean);四个组件 checkout 均精确处于被 pin 的 SHA 且无脏文件(agent-intent 7f3ff6a、native be556f4、squad 95b387d、backend bd5c0f3)。
- `gofmt -l intent/ cmd/`:无输出(格式干净)。
- `go test`:**NOT_RUN(环境不可编译)**。agent-intent/go.mod 的 `replace error-code-system => ./deps/error-code-system` 所指嵌套子模块在本工作区任何层级都未物化(superrepo、agent-intent、backend 三处的 error-code-system/logging-system 目录均为空挂载点),且相应 commit 对象不存在于任何本地对象库;单元约束禁止 `git submodule init/update`、禁止网络。`GOWORK=off go test ./intent/...` 如预期在包 setup 阶段失败(reading deps/error-code-system/go.mod: no such file)。诚实出口:本报告的测试结论为读码+探针+追踪,非重放。
- 否定判定探针(negatedAt/inNonNegatingBie 逐字提取、21 组用例):全部 corpus 否定序列判否定;「不断」「无偿」不触发误否定;「不要马上重启」(间隔 2 字)判否定、「不要立刻马上重启」(间隔 3 字)不判(窗口语义);「特别的,重启任务」跨标点正确豁免;「任务别名重启」豁免成功;「判别任务进度」被误判否定(见 F1)。
## 评审发现
### Blocker
无。
### Major
无。
### Minor
- **F1(recall 损失,fail-safe 方向)**:「别」字白名单 `nonNegatingBie`(intent/registry.go:211-214)是封闭列表,「判别」「辨识」「作别」等含「别」的常用词不在列。探针实测「判别任务进度」→「进度」命中被 `negatedAt`(registry.go:263-275)判为否定 → status_query 以 `negated_match` 排除 → 回落 fallback。方向是安全的(少执行而非误执行),但对该输入是误回落。建议后续把「别」的否定判定改为右边界词表或按「别+动词」结构判断。
- **F2(残余否定词缺口,符合既定设计范围)**:token 表(registry.go:204)无「无须」「勿」「先不」「暂不」「不必」等变体,如「无须重启任务」仍会正向命中 restart。任务卡设计决定明确固定了这 7 个词,故不算实施偏差;作为残余风险记录:negation_reference 类的其他说法仍可能误执行。
### Notes
- **N1(tie-break 只认“有效”槽位)**:`resolveTie`(intent/recognizer.go:152-185)用 `len(slots)` 计数,invalid 捕获(如多候选值)不计。因此「你好,取消任务 42 和任务 43」这类输入中 cancel 的 run_id 无效 → filled=0,与寒暄同为 0 时按 priority(greet=10 > cancel=0)裁决到寒暄,动作侧的 clarify 丢失。仅在同分且优先级不同时可达,且寒暄无副作用、动作未启动,属可接受行为选择;若希望「带槽位信号」包含无效捕获信号,需另行设计。
- **N2(fail-closed 误伤面)**:`hasInnerSeparator`(slots.go:221)/`captureTruncated`(slots.go:201)对合法值也拦截——如主题「和平与发展」会因内部分隔词判多候选、值后紧跟 ASCII 词字符即判截断,一律 clarify 而非 start。与「无效一律 clarify」的设计一致,代价是 clarify 率上升(metrics 中 wrong_slot clarify 7/7=1.0),方向安全。
- **N3(测试夹具标签失真,表面问题)**:cmd/intentctl/main_test.go:82-85 把 record_only 的 miss 用例从「不用查进度了」换成「进度条的颜色真好看」是必要的(旧输入修复后正确回落,不再产生 miss 计数;新输入在 testSpecJSON 下确会命中 status_query,仍覆盖 miss 记账路径,断言 MisExecutions==1 成立),但 category 仍标 `negation_reference`,语义已不属否定类——建议下次触及时改名或换语料类别。
- **N4(验收证据等级)**:mis_exec=0 / 37/37 的验收来自 intent@7f3ff6a 内提交的 evaluation/metrics.json 与 BASELINE.md(作者证据,SELF 级)。本工作区无法重放(见上)。数字内部一致性已人工复核:各类 passed 7+8+7+8+7=37、clarify 5+0+5+0+7=17、fallback 3+1+0+7+0=11、top1 11+7+3+3+7=31/31、mis_exec 全 0,与 overall 行及 I2 段(TP=30/FP=0/FN=0,较基线同向 +1)全部吻合;gen_agree 5/8 未动(3 例 gen_agree 分歧按范围未处理,BASELINE.md 如实保留)。CI(`.gitea/workflows/ci.yaml` @ PR HEAD)是测试通过的权威源,合入门禁照常。
- **N5(单元包描述与实际 diff 的 pin 基线不符,仅描述性)**:包内写 squad b6d8763→95b387d、backend 330c1a3→bd5c0f3;实际 vs main 为 squad a5afe9c8→95b387d、backend 8dc91c17→8031855e→bd5c0f3(两个 backend tip 树完全相同,bd5c0f3 非 8031855e 的后代而是同树重提交)。终态与包目标一致,无内容影响。
- **N6(任务卡整理欠账)**:`.task/feature-recall-misexec-fixes.md` 的「Agent 实施收据」表仍是全 NOT_RUN/待定义,而「合入与清理」已记录实施收据;需求分析/阶段记录等模板段未填。卡片债务,非阻塞;Review 记录待本评审随 PR 流程回填。
- **N7(正面确认,范围)**:变更严格限于三个修复+pin 级联+任务卡:intent 单提交共 9 文件(boundary.go 为纯重构抽取 `boundedAt`,语义等价);squad@95b387d、native@be556f4 两提交均只改 gitlink;嵌套 pin 全链一致(native→intent@7f3ff6a;squad→intent@7f3ff6a+native@be556f4;backend→intent@7f3ff6a+native@be556f4+squad@95b387d,与 superrepo 四个 gitlink 完全对应)。语料文件 recall_corpus.json 未改动(验收基准未被移动)。
- **N8(正面确认,正确性要点)**:否定为逐次出现判断,同词条存在任一未否定出现即正向(TestNegationCountsUnnegatedOccurrence/TestNegatedPatternHit 覆盖了否定+未否定同词双情形);守卫域豁免否定(registry.go:236、254,宁可多拦,TestGuardIgnoresNegation);excludes_any/includes_all 语义未动;r2 回归测试改名并反转断言是对旧契约的有意修订且注释说明清楚;`resolveTie` 确定性(filled 降序→priority 降序→合并序,胜者须压过组内全体,平 priority 仍 ambiguous),仅在胜出时原位重排 hits,Explain/Recognize 双路径一致,I5 行为保留(TestRecognizeTieBreakPrefersPriority 未改);槽位新检查在 format/enum/validator 之前(slots.go:186-198),空白归一化后进 missing(slots.go:123),overlap 检测未动;misexec_test.go 逐条 pin 了 6 次误执行序列与 4 个错槽位序列(含 start+clarify 命令断言),非否定输入行为回归用例齐备。
- **N9(字节级安全性)**:`termHit` 以 `i+1+j` 推进的重叠出现扫描与 `negatedAt` 的 `DecodeLastRuneInString` 回退均为 rune 安全;否定判定在折叠文本上进行,否定词为 CJK,大小写折叠/NFKC 不影响。
## 结论
三个修复(否定感知匹配、specificity tie-break、槽位 fail-closed)实现与任务卡设计决定一致,验收序列全部有单元级 pin,pin 级联自洽,范围无外溢。发现集中在启发式边界的已记录取舍(F1/F2)与非阻塞整理项;无阻断合入的问题。测试在本评审工作区因嵌套依赖未物化而 NOT_RUN,以探针与读码代替,合入前按门禁以 CI 为准。
REVIEW VERDICT: APPROVE
+6 -3
View File
@@ -60,8 +60,9 @@ done
# ── 1. 构建 ─────────────────────────────────────────────
# RDM 源码经 git archive 从本地对象导出(不建 worktree、不触远端——Gitea 抖动时
# submodule 克隆会挂死);pins/* 从 RDM 主检出的已物化 submodule 快照拷贝(cp 排除
# .git,39MB 级),构建输入由此固定为 RDM origin/main + 其 gitlink 指向的本地内容。
# submodule 克隆会挂死);pins/* 从 RDM 主检出的已物化 submodule 快照拷贝(tar 排除
# .git;含 go.mod replace 引用的全部 pin,不仅 backend),构建输入由此固定为
# RDM origin/main + 其 gitlink 指向的本地内容。
log "build RDM binary (archive origin/main + snapshot pins from local checkout)"
RDM_WT="$OUT/rdm-src"
RDM_SHA=$(git -C "$RDM_REPO" rev-parse origin/main) || die "no origin/main in RDM repo"
@@ -75,7 +76,9 @@ PIN_SHA_WANT=$(git -C "$RDM_REPO" ls-tree "$RDM_SHA" pins/agent-system-backend |
[ "$PIN_SHA_MAIN" = "$PIN_SHA_WANT" ] \
|| die "RDM checkout pins/backend ($PIN_SHA_MAIN) != origin/main pin ($PIN_SHA_WANT); git -C $RDM_REPO submodule update first"
mkdir -p "$RDM_WT/pins"
tar -C "$RDM_REPO/pins" --exclude=.git -cf - agent-system-backend | tar -C "$RDM_WT/pins" -xf -
# 快照全部 pins/*(go.mod replace 依赖 agent-document-renderer、agent-squad 等,
# 仅拷贝 backend 会导致 RDM 构建缺 pin);--exclude=.git 保持不变。
tar -C "$RDM_REPO/pins" --exclude=.git -cf - . | tar -C "$RDM_WT/pins" -xf -
cleanup() {
[ -n "${API_PID:-}" ] && { kill "$API_PID" 2>/dev/null || true; wait "$API_PID" 2>/dev/null || true; }
# Preview runs directly as one owned Vite process. Verify its command before