Why: establish the authorized task workspace and its initial baseline.
Evidence: preserved precheck receipt in .task/feature-agent-quality-phase01.md; R0=1ebd46edc5239d0d87418e9a8dd9aa225955e1dd.
Impact: task metadata only; the feature workspace starts at 1ebd46edc5.
Refs: .task/feature-agent-quality-phase01.md
Why: The 2026-10-09 diagnosis showed behavior-surface regressions (gate
deadlocks, silently broken lint, wrong artifact verdicts) passing every unit
test because no task-level check ran the real model. User ruling 6 approved
making before/after live-eval evidence a delivery requirement for prompt,
skill, tool-description, context-assembly and gate changes.
Evidence: SELF - clause drafted to reference modules/backend/eval/live and the
GATE/REPLAY/SELF sourcing already defined in AGENTS.md; task card for
feature/agent-quality-phase01 filled with the P0+P1 scope and requirements.
Impact: policy/delivery.md gains the 智能体行为面的真实验证 subsection;
NOT_RUN stays legal but must be explicit. Phase01 task card records scope,
requirements R1-R4 and the before/after eval protocol.
Refs: modules/documents proposal 2026-10-09-agent-quality-improvement-plan.md (ruling 6).
Why: Backend carries the seven quality-gate repairs with the eval/live
harness; documents carries the improvement-plan main record with the six user
rulings. The Dev combination must test these heads together.
Evidence: SELF - gitlink pointers match the pushed branch heads; make test
recorded on the task card after this commit.
Impact: Combined test surface now includes the phase01 backend and documents
content; no behavior change from the pointers alone.
Refs: feature/agent-quality-phase01 (backend), feature/agent-quality-plan (documents).
Why: The previous pointer predated the eval/live package commit; the
combination must test the full phase01 backend branch.
Evidence: SELF - gitlink equals backend feature/agent-quality-phase01 head.
Impact: Pointer only.
Why: The lint skills-root injection moved from the app root to the runner,
keeping the architecture ratchet green; the Dev combination must test that head.
Evidence: SELF - gitlink equals backend feature/agent-quality-phase01 head;
make test rerun recorded on the task card.
Impact: Pointer only.
Why: The backend branch gained the user-scope model capture fix, and the
live eval produced the before-baseline report on remote main (1ebd46e) that
the task card and the improvement-plan record reference as evidence.
Evidence: SELF - report .task/phase01-eval/before-main-1ebd46e.json (model
MiniMax-M2.7-highspeed, 4 golden tasks, 17 min run): reproduces the diagnosed
failure chain on the pre-fix build - readability dead-loop x4 on two tasks,
9 full regens with 0 original-material reads in plan-followup, 8 unverified
percentages, forged-marker false positive, internal check text leaked to the
user.
Impact: Report committed for the after-comparison once this PR merges; gitlink
points at the harness-inclusive backend head.
Why: Ruling 6 requires real-model before/after evidence. The after run (fixed
build b68833f, same golden tasks) proves the repairs landed: lint now actually
executes (real layout/lead findings appear vs silent empties before), the
pm-tool readability dead-loop is gone (4/0 -> 0), plan-followup reads original
materials 0 -> 49 times. Remaining failures are the documented P5/P2 scope
(rename-revision loops via scorer semantics, unverified numbers, verifier
parse hiccups), not regressions.
Evidence: SELF - report .task/phase01-eval/after-phase01-b68833f.json +
evidence row (1749s run); comparison table in the card.
Impact: PR #277 now carries the complete before/after pair for review.
weichen
merged commit 4deddf458d into main2026-10-09 14:13:07 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
摘要
智能体工作流质量改进首个增量（P0 评测基线 + P1 缺陷修复）。主记录与六项裁定：
Documents #51（
2026-10-09-agent-quality-improvement-plan.md）。组件 PR：
渲染格式与伪造标记同形、标题误报、python3 命令、检查提醒措辞、rune 截断）
eval/live真实模型评测基座本 PR：
前后对照，无条件跑时显式 NOT_RUN）
.task/feature-agent-quality-phase01.md（R1–R4、证据行、结果与剩余项）.task/phase01-eval/before-main-1ebd46e.jsonbefore 基线（旧 main 1ebd46e，真实模型，两轮）
（两轮数字为 首跑/复跑，失败链稳定复现。）
Evidence
合入顺序与剩余项
Backend #111 → Documents #51 → 本 PR。after 对照在修复版部署后按同一任务集补跑
（已知未解：改名重登记仍会死循环 → 裁定 4 + P5 处理；tavily 播种后需重启才注册 → 另行修复）。