总账INFRA LEDGER · guiming
2026-07-22 构建快照投影 · 真源在仓
LESSONS × ERRORS

教训库

教训(各 run FINAL_REPORT 提取)与错误台账合并浏览;主视图按复发聚类,同一坑出现两次即显形。

数据 as-of 2026-07-22

复发聚类(26 簇)

教训 109 + 错误台账 165 · 聚类键=recurrence_tag/BUG-ERR 族

复发簇 · untagged

91 次
教训2026-07-19

「生成器成第二真源」风险(预防性)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/design/D1_FINAL.md

教训2026-07-19

「GENERATED 区被手改」风险,缓解=banner+pre-commit(预防性)

展开

根因:

对策:banner+pre-commit 缓解

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/design/D1_FINAL.md

教训2026-07-19

「gate 无消费者、静默失能(invocation:manual 综合征)」(预防性)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/design/D1_FINAL.md

教训2026-07-19

「casual 层再膨胀:每加 provider 往表加列加行」(预防性)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/design/D1_FINAL.md

教训2026-07-19

残差:14 域 status rollup/gate 计数/cron 面/决策摘要/过时资产五面真源未接;max_staleness_days 只写不启用;图节点名校验器未写

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/design/D3_FINAL.md

教训2026-07-19

LEDGER:188 教训:infra-newroot 块缺注释导致未装

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/design/D4_FINAL.md

教训2026-07-19

残差:14 域 status/gate/cron/决策摘要五面真源未接、图语义正确性无第二层校验

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/design/D4_FINAL.md

教训2026-07-19

额度墙中断:W-C 完稿未及自报、F2 重派实例失控覆写设计文件

展开

根因:额度墙中断后重派实例无原 session 约束

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r002_site_overhaul_20260719/reports/FINAL_REPORT.md

教训2026-07-19

台账结构缺口:DECISION_LEDGER_BACKFILL.md 主表缺 r010/r011 两节

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r002_site_overhaul_20260719/reports/FINAL_REPORT.md

教训2026-07-19

误派已有产物的新写任务:「给仓库的信」被当新任务派 Fable 写作,用户叫停

展开

根因:

对策:「写 X」任务先做 existence check(已入记忆 feedback_info_first_phase_planning)

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r002_site_overhaul_20260719/reports/FINAL_REPORT.md

教训2026-07-19

残差:地图/站点数据保鲜 cron 未装;claim ceiling=bounded

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r002_site_overhaul_20260719/reports/FINAL_REPORT.md

教训2026-07-20

commit_wrapper --message 不带路径会吞整个 index(fd58401 误吞 994 文件他人工作)

展开

根因:

对策:本仓手动 commit 必须 git add <paths> && git commit -- <paths>

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r003_site_remediation_20260720/reports/FINAL_REPORT.md

教训2026-07-20

K3 max_turns=60 陷阱:大图区审查 60 turns 只够读完

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r003_site_remediation_20260720/reports/FINAL_REPORT.md

教训2026-07-20

残差:R6 剩余 74 条决策联动字段空;kimi-ollama 通道空响应故障

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r003_site_remediation_20260720/reports/FINAL_REPORT.md

教训2026-07-21

FINAL_REPORT 尚未生成,V6/V7/V12/V13 四条验收锚无载体

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r004_site_content_expansion_20260721/impl/K3_FINAL_REVIEW.md

教训2026-07-21

检查路径弯路 1:decisions.html 首查 relates_systems grep 仅 1 命中,一度疑似字段未渲染(后证非缺陷)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r004_site_content_expansion_20260721/impl/K3_FINAL_REVIEW.md

教训2026-07-21

检查路径弯路 2:source 层按 id: 正则切分误把 relates_systems 数组值当条目(115 vs 99)(后证非缺陷)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r004_site_content_expansion_20260721/impl/K3_FINAL_REVIEW.md

教训2026-07-21

口径差异:V8 措辞「74 条」与 LINK74_STATE「总条目 99」不一致

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r004_site_content_expansion_20260721/impl/K3_FINAL_REVIEW.md

教训2026-07-21

F-T9-1:台账 BF-26 与站投影「12 题全未裁」过时(Q7/Q8 已裁)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r004_site_content_expansion_20260721/reports/FINAL_REPORT.md

教训2026-07-21

遗留清单 L-1~L-12(12 项待用户裁定,含设计哲学 10 题、波次 1 启动时机等)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r004_site_content_expansion_20260721/reports/FINAL_REPORT.md

教训2026-07-21

publish cron 自 07-20 起连败:publish.mjs 裸调 node,cron 无 nvm PATH

展开

根因:cron 极简 PATH 无 nvm

对策:cron 环境显式钉 PATH(nvm 具体版本 bin)

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r005_dashboard_redesign_20260721/reports/FINAL_REPORT.md

教训2026-07-21

BSD script 伪 TTY 在 headless 下不可用:stdin 为 socket 时 tcgetattr 直接失败

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r005_dashboard_redesign_20260721/reports/FINAL_REPORT.md

教训2026-07-21

worker 断点续棒竞态:首棒死后自动重试抢在 charter 补丁落盘前 21 秒启动、拿到旧 prompt

展开

根因:

对策:charter 补丁必须先于重试窗口

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r005_dashboard_redesign_20260721/reports/FINAL_REPORT.md

教训2026-07-21

tier 名必须带档位后缀:session_dispatch --channel claude-kimi 报 unknown_tier

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r005_dashboard_redesign_20260721/reports/FINAL_REPORT.md

教训2026-07-21

遗留 R-1~R-5(沙箱 SIGTERM、密码门防护级别、巡检 cron 首跑待自证等)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r005_dashboard_redesign_20260721/reports/FINAL_REPORT.md

教训2026-07-19

channel 健康标签把 scope 冷却投影到传输名造成「OpenCode 没额度」误读

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/synthesis/FINAL_REPORT.md

教训2026-07-19

合规脚本正则把历史叙述误判为指针反模式(被交叉复审官纠正)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/synthesis/FINAL_REPORT.md

教训2026-07-19

cd 在复合命令中可能被拦而工作目录跨调用持久化,两者叠加产生了三次无害但迷惑的 ENOENT 部署失败

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/synthesis/FINAL_REPORT.md

教训2026-07-11

用户仍在手工调度。调研结果、残差、priority 和 session prompt 没有连成一个 selector

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/design/FINAL_MIGRATION_OPERATING_PLAN.md

教训2026-07-19

残留 MINOR:R3「fallback 26/18 未本轮逐行重放」、R4「约 70 commit 子集归属未逐一隔离」(均判不 BLOCKING)

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/audit_20260719_recent_refactor_sessions/FINAL_CLAIM_AUDIT.md

教训2026-07-19

唯一 wrong 数(c736c471 116)已被报告订正为 106

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/audit_20260719_recent_refactor_sessions/FINAL_CLAIM_AUDIT.md

教训2026-07-13

残差登记:r008-R1 凌晨 30Gi 波动根因未坐实;r008-R2 diff-temp 泄漏源未定谳;r008-R3 T1463 归属分歧

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/r008_20260713_status_synthesis_and_relaunch/reports/FINAL_REPORT_r008.md

教训2026-07-14

样例教训已折入:worker 必真跑测试+主编排复验贴 summary

展开

根因:

对策:worker 必真跑测试+主编排复验贴 summary

adhoc_jobs/context_infra_migration_20260709/runs/r010_20260714_d1a_landing_and_pilot_units/reports/FINAL_REPORT_r010.md

教训2026-07-14

残差登记:r010-R1 error_ledger 无轮转;r010-R2 新根史首个 message 越述 commit;r010-R3 sub-agent 回执延迟 ~2.5h 通道根因未查;r010-R4 旧根 settings.json 三个死 hook 槽静默吞越界

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/r010_20260714_d1a_landing_and_pilot_units/reports/FINAL_REPORT_r010.md

教训2026-07-14

Kimi 报 DONE 未真跑测试(抓住);GLM 自迭代后空转 15 分钟(接管终止)

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/r010_20260714_d1a_landing_and_pilot_units/reports/FINAL_REPORT_r010.md

教训2026-07-14

11 条 findings:5 处当场修复,4 条入残差,2 条安全确认项

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/r011_20260714_injection_v2_hygiene_and_report/reports/FINAL_REPORT_r011.md

教训2026-07-14

残差登记:r011-R1 O1 oracle 判据欠拟;r011-R2 50d519c 夹带 carry 的 rollback 语义;r011-R3 sweep 白名单 check-ignore 双确认未做;r011-R4 TMPDIR 报数只扫顶层

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/r011_20260714_injection_v2_hygiene_and_report/reports/FINAL_REPORT_r011.md

教训2026-07-14

r011 的 claims 双 schema 教训直接喂它的 schema 设计

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/r011_20260714_injection_v2_hygiene_and_report/reports/FINAL_REPORT_r011.md

教训2026-07-16

残差登记:r012-R1 patrol_reports 留存策略;r012-R2 旧根→新根反向波及六边未消除;r012-R3 route_hint 依赖 INDEX triggers 人工维护;r012-R4 eval case1_A DNF 未重试

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/r012_20260716_injection_v3_orchestrator_isolation/reports/FINAL_REPORT_r012.md

教训2026-07-16

I1(High)「S7 引擎切换未上线,生产提交仍走 Level 1 per-session index」

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/r012_20260716_injection_v3_orchestrator_isolation/units/U-EVAL-AB/results/artifacts_case2_audit_A/FINAL_EVALUATION_REPORT.md

教训2026-07-16

I4(High)「Stop hook synthetic staging 反复 pathspec 失败,主工作区提交受阻,积压形成」

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/r012_20260716_injection_v3_orchestrator_isolation/units/U-EVAL-AB/results/artifacts_case2_audit_A/FINAL_EVALUATION_REPORT.md

教训2026-07-16

O3(High)「staged 区含 398K 删除,可能涉及整目录归档,需人工确认是否为预期清理」

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/r012_20260716_injection_v3_orchestrator_isolation/units/U-EVAL-AB/results/artifacts_case2_audit_A/FINAL_EVALUATION_REPORT.md

教训2026-07-16

I2/I3/I5/I6/O1/O2(Medium/Low):mirror ref 迁移未完成、PostToolUse hook 未挂载、pending_merges 未解决冲突、main 领先 origin 7583 提交、audit.log 停留 04-16、crontab 文档与实际有后向差异

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/r012_20260716_injection_v3_orchestrator_isolation/units/U-EVAL-AB/results/artifacts_case2_audit_A/FINAL_EVALUATION_REPORT.md

教训2026-07-16

残留项 R1-R12「全部给出为什么没做+如何再做」

展开

根因:

对策:

adhoc_jobs/context_infra_migration_20260709/runs/r012_20260716_injection_v3_orchestrator_isolation/units/U-EVAL-AB/results/artifacts_case2_audit_A/FINAL_EVALUATION_REPORT.md

教训2026-07-11

停摆事故:03:31 r2/critic-a 转 idle 仅发 idle 通知(不唤醒主线),主线 03:40-06:51 停摆 ~3h

展开

根因:

对策:收口必须发完成消息(已入所有 brief)

adhoc_jobs/context_infra_base_tooling_buildout_20260615/methodology_landing_campaign_20260711/reports/FINAL_REPORT.md

教训2026-07-11

HG-INCIDENT-rmrf:p3 前任 Fable 执行 rm -rf 删了不可复原的 07-02 S6 dogfood 包(untracked,git 无副本)

展开

根因:

对策:把「untracked 且被 INDEX/文档引用的 dogfood/证据包」纳入 git_guard 档 C 保护清单

adhoc_jobs/context_infra_base_tooling_buildout_20260615/methodology_landing_campaign_20260711/reports/FINAL_REPORT.md

教训2026-07-12

已确认非阻塞问题:URA validator 不强制 locator 绝对路径约束;旧结论 294 因动态记录源变化失效

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/methodology_landing_campaign_20260711/audits/session_bb3655a3_prompt_injection_20260711/tests/FINAL_INDEPENDENT_REVIEW.md

教训2026-07-12

需保留风险:dispatch preflight 与 worker read 之间有 TOCTOU 窗口;receipt 结构化 read evidence 为空

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/methodology_landing_campaign_20260711/audits/session_bb3655a3_prompt_injection_20260711/tests/FINAL_INDEPENDENT_REVIEW.md

教训2026-07-11

orchestrator__M2 无 ROLLOUT_REPORT(结构性缺件)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/methodology_landing_campaign_20260711/impl/rollout/FINAL_EVAL.md

教训2026-07-11

landing_gate_consumer_assert.py 行为探针每次运行会在 contract/_probe_tmp/ 留一个 plan.yaml——断言脚本反复跑会累积残留

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/methodology_landing_campaign_20260711/impl/rollout/FINAL_EVAL.md

教训2026-07-04

F1 判决挤占(5 跑 3 中):regulator session 被 repo phase hook 干扰

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/production_readiness_20260704/FINAL_REPORT.md

教训2026-07-04

残差:BT-PROD-F2/F3、15 gate 迁移债、判断 Agent 校准样本积累

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/production_readiness_20260704/FINAL_REPORT.md

教训2026-07-10

深水区遗留(已记录未做):Stop hook 快慢两段式重构(本轮=提速+提额,非终态);生成侧无界家族三件(receipts.jsonl 273MB、TODO.md.bak rotation、Run5 scoreboard ~600 untracked)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/runs/r009_20260710_state_gap_quick_landing/FINAL_REPORT.md

教训2026-07-10

main 领先 origin 6397+ commits(push 停在 05-17)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/runs/r009_20260710_state_gap_quick_landing/FINAL_REPORT.md

教训2026-07-10

guardian 健康报告消费面无消费者

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/runs/r009_20260710_state_gap_quick_landing/FINAL_REPORT.md

教训2026-07-10

shadow 99.8% divergent

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/runs/r009_20260710_state_gap_quick_landing/FINAL_REPORT.md

教训2026-07-08

仅一句通用声明:Residuals are registered in this report and preserved in current generated status surfaces.(残差,无具体条目)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/runs/r007_20260708_testsys_buildout_completion/reports/FINAL_REPORT.md

教训2026-07-05

写作本报告的 70 分钟里台账又涨了 2,043 行(133,572→135,615),问题正在实时恶化(hook timeout 根因现场实证)

展开

根因:hook timeout 导致台账无界增长

对策:

adhoc_jobs/test_system_landing_methodology_20260704/reports/FINAL_REPORT.md

教训2026-07-05

测试抓过吗:没有,且结构上不可能抓到——route_on_prompt 27 个测试+rr4 5 个测试全部用空台账(亲手把致病自变量归零)

展开

根因:

对策:

adhoc_jobs/test_system_landing_methodology_20260704/reports/FINAL_REPORT.md

教训2026-07-05

Run5 两点不达:其一,转述的「跑完了、推到完全 production」超它自报两级;其二,「实质关闭」停在协调者自查档,且关闭不回写记分牌证据源

展开

根因:

对策:

adhoc_jobs/test_system_landing_methodology_20260704/reports/FINAL_REPORT.md

教训2026-07-05

两起本 session 现场事故(检针自命中、谓词分级偏好存在性)已作为一手根因证据入案

展开

根因:

对策:

adhoc_jobs/test_system_landing_methodology_20260704/reports/FINAL_REPORT.md

教训2026-07-05

首跑在共享 receipts 台账的排他锁上挂起 10 分钟(多 campaign 并发下锁队列已到分钟级),改用隔离台账 env 后秒过

展开

根因:

对策:

adhoc_jobs/test_system_landing_methodology_20260704/reports/FINAL_REPORT.md

教训2026-07-05

残差:Run5/i2p round6 在飞;记分牌盲点其余 11 域未逐格审;「有机≈0」是 7 天窗保守快照

展开

根因:

对策:

adhoc_jobs/test_system_landing_methodology_20260704/reports/FINAL_REPORT.md

教训2026-07-10

仪器双 fixture 校准常规化(C,BT r003 仪器伪影占 unknown 六成的教训):任何确定层仪器输出进表前过一次 known-good + known-bad 双 fixture

展开

根因:

对策:确定层仪器输出进表前过 known-good + known-bad 双 fixture

adhoc_jobs/systems_consolidation_20260705/collection_first_methodology_20260710/FINAL_METHODOLOGY.md

教训2026-07-05

3 处 stale 门(orchestrator 两选一 06-30 已拍板仍标 await_user、Run1 merge 门已事实解决、git crontab 已装仍被记「待装」)

展开

根因:

对策:

adhoc_jobs/systems_consolidation_20260705/CONSOLIDATION_FINAL_REPORT.md

教训2026-07-05

6 处状态谎标

展开

根因:

对策:

adhoc_jobs/systems_consolidation_20260705/CONSOLIDATION_FINAL_REPORT.md

教训2026-07-05

MC3 实抓 64 处弯引号转写损伤并修复至字节级 EXACT

展开

根因:

对策:

adhoc_jobs/systems_consolidation_20260705/CONSOLIDATION_FINAL_REPORT.md

教训2026-07-05

遇到的问题如实记录:MC3 FLAG 三类反例与修复、miner 中间文件碰撞事故、phase 回执窗口摩擦、gate 首跑 FLAG 补标准段

展开

根因:

对策:

adhoc_jobs/systems_consolidation_20260705/CONSOLIDATION_FINAL_REPORT.md

教训2026-07-05

诚实边界:14 系统全量复测未做(记分牌黄区非盲区);25 票推荐≠正确

展开

根因:

对策:

adhoc_jobs/systems_consolidation_20260705/CONSOLIDATION_FINAL_REPORT.md

教训2026-07-05

机械底盘强,失手在决策生成层(无类型学过滤)与执行衔接层(「接手」无属主)。七条教训入负面清单,其中 L4(批准必须带启动三元组)已写进本轮执行计划每一项

展开

根因:

对策:L4:批准必须带启动三元组

adhoc_jobs/systems_consolidation_20260705/R3_FINAL_REPORT.md

教训2026-07-05

残差:S7 shadow 发散未重跑;DDP 面 D-1..D-5 执行痕迹未逐项核实;A5 底表 3 行未核实

展开

根因:

对策:

adhoc_jobs/systems_consolidation_20260705/R3_FINAL_REPORT.md

教训2026-07-05

「本轮只读,不修复任何被查系统」「T1237-T1262 是后续修复入口,不代表缺陷已解决」(性质克制的残差声明)

展开

根因:

对策:

adhoc_jobs/systems_consolidation_20260705/runs/r002_execution_quality_sweep/FINAL_EQS_REPORT.md

教训2026-07-05

「未安装 cron/launchd,符合 prompt 纪律」「TODO 候选注册仍留人审,未自动 intake」「历史 buildout RD6 backlog 仍存在」

展开

根因:

对策:

adhoc_jobs/systems_consolidation_20260705/runs/r007_w2a_bugtrace_ddp/FINAL_REPORT.md

教训2026-07-20

按域残差注册表 CAND-0007~0014(8 条):史实域缺 skill 层受控入口;error_ledger 对独立 session 不可达;open-ended×headless 超时致不可观测;receipts path 归因只覆盖 rules/skills/;taxonomy 载体两处遗漏;静态 grep 对运行时渲染站系统性低估图数(等)

展开

根因:

对策:

adhoc_jobs/methodology_system_buildout_20260719/eval/P5_FINAL_REPORT.md

教训2026-07-20

开线记账:version-manager 校核未做;record.py "null" 字面串缺陷;B3 未落地 2 锚 delta 提案流转无回执

展开

根因:

对策:

adhoc_jobs/methodology_system_buildout_20260719/eval/P5_FINAL_REPORT.md

教训2026-07-20

§5 残差登记 7 条:matrix 头行计数 stale;S3 批2/批3 未派;receipts path 归因不对称;共享索引 457 文件 staged-delete 陈旧;methodology INDEX 6 行未迁移

展开

根因:

对策:

adhoc_jobs/methodology_system_buildout_20260719/FINAL_REPORT.md

教训2026-07-04

V1-V6 验证义务未跑之前,不得声称系统有效或可直接投产(claim ceiling)

展开

根因:

对策:

adhoc_jobs/requirement_decomposition_context_isolation_20260704/reports/FINAL_REPORT_20260704.md

教训2026-07-04

残差:书籍证据全部来自蒸馏产物;隔离三变量独立性未受控验证;G9 进程安全无基建映射

展开

根因:

对策:

adhoc_jobs/requirement_decomposition_context_isolation_20260704/reports/FINAL_REPORT_20260704.md

教训2026-07-05

receipt 台账锁护航(四次独立复现→登记 BUG-R5-13→T1113)

展开

根因:

对策:

adhoc_jobs/requirement_decomposition_context_isolation_20260704/reports/FINAL_REPORT_R002_20260704.md

教训2026-07-05

phase driver fail-open(登记 BUG-R5-12→T1108)

展开

根因:

对策:

adhoc_jobs/requirement_decomposition_context_isolation_20260704/reports/FINAL_REPORT_R002_20260704.md

教训2026-07-05

V4 跑臂四轮通道迭代(每轮有机械诊断,最终 6/6)

展开

根因:

对策:

adhoc_jobs/requirement_decomposition_context_isolation_20260704/reports/FINAL_REPORT_R002_20260704.md

教训2026-07-10

主 session context 已被上一轮污染(混乱中间产物/试错痕迹在场)→ 外移从默认升为强制(判断质量不被走过的弯路带偏)

展开

根因:

对策:判断外移 clean session 从默认升为强制

adhoc_jobs/diagram_learning_systems_20260701/optimization_clean_session_judgment_20260710/FINAL_ADJUDICATION.md

教训2026-07-10

tools/mermaid_diagram_check:本轮不动(SOLID_DECISION Family-2 教训+既有裁定)

展开

根因:

对策:

adhoc_jobs/diagram_learning_systems_20260701/optimization_recursive_subsystem_20260710/FINAL_ADJUDICATION.md

教训2026-07-10

deliverable-first 提醒(库内 07-08 跑偏教训):第 6 步的重画 dogfood 是本轮优化的真验收物,勿在名册工具化上兔子洞

展开

根因:

对策:deliverable-first:先保真验收物,勿兔子洞名册工具化

adhoc_jobs/diagram_learning_systems_20260701/optimization_recursive_subsystem_20260710/FINAL_ADJUDICATION.md

教训2026-07-01

先方法论后工具——这是本仓画靶射箭教训的正面应用

展开

根因:

对策:

adhoc_jobs/diagram_learning_systems_20260701/reports/FINAL_REPORT.md

教训2026-07-01

残差:两系统本体是设计态(卡片库无真卡、方法论 skill 未写);手绘照片→结构 diff 保真度未验;绘图效应外推到复杂概念图未验

展开

根因:

对策:

adhoc_jobs/diagram_learning_systems_20260701/reports/FINAL_REPORT.md

教训2026-07-13

仅指向 RESIDUAL_REGISTRY.json 机器可读真源(当前 9 项全部保持),报告正文未展开叙述(指针型残差)

展开

根因:

对策:

adhoc_jobs/design_doc_protocol_loop_codex_20260501/runs/r007_20260711_clean_context_runtime/reports/FINAL_REPORT.md

教训2026-07-11

Residual:P1 只证明来源覆盖与当前实现事实边界,不证明需求模型、设计或实现已通过(范围声明)

展开

根因:

对策:

adhoc_jobs/design_doc_protocol_loop_codex_20260501/runs/r007_20260711_clean_context_runtime/reviews/P1_SOURCE_COVERAGE_FINAL_REVIEW.md

教训2026-07-19

EvidenceAnchor.kind 英文短词枚举(file/session)与站点固有词(.file-button/sessionStorage)子串碰撞,断言面被污染

展开

根因:英文短词在英文语料子串匹配中天然高碰撞

对策:注入宿主环境做断言/匹配用的枚举值避免英文短词;用二字中文枚举(会话/文件/榜面/提交/链接/事件,<4 字符免进禁止集)或长尾命名

career/runs/r007_20260719_orchestration/website_impl_state.md

教训2026-07-19

career r007 archiver 归档根路径多上溯一级落仓库顶层(WAVE_S3 验收抓到当场修复),但 career 线无 bug 台账登记面,bug 只活在 run state 行里

展开

根因:career 线缺 error_ledger/bug 台账机制,bug 无登记落点

对策:

career/runs/r007_20260719_orchestration/website_impl_state.md

教训2026-07-05

两条 07-05 已裁纪律四登记面零命中:Q22 session 收尾必注册残差(收口谓词加注册回执要求+纪律成文,方案 B+A 已裁,Stop hook 方案已否决);自决判据三层模式(有引导时自主为主/确实不确定且细节关键才问人/不关键且多方案则自动测试择优,P-008 扩展通过)

展开

根因:

对策:

adhoc_jobs/systems_consolidation_20260705/USER_DECISION_SUMMARY.md

复发簇 · auto_regulator_runner:missing_task_id:no_tier:missing_task_id

32 次
错误台账unresolved2026-07-04

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-04

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-04

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-04

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-04

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-04

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-05

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-05

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-05

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-05

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-05

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-05

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-05

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-05

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-06

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-07

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-07

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-07

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-09

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-09

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-09

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-09

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-09

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-11

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-11

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-11

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-12

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-12

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-12

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-12

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-19

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-19

regulator_runner missing_task_id: regulator_runner requires TODO_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/missing_task_id/missing_task_id

tools/infra_core/regulator_runner.py:run_regulator

复发簇 · auto_regulator_runner:main_environment_required:no_tier:main_environment_required

14 次
错误台账unresolved2026-07-04

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.; receipt_id backfilled by tools/error_ledger/triage.py: TR-126970 score=100 reason=error_code+chain_step+domain+time<=585s

orchestration_failure · router_soft · regulator_runner/T1090-W1GD/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-04

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.; receipt_id backfilled by tools/error_ledger/triage.py: TR-126970 score=140 reason=task_id+error_code+chain_step+domain+time<=0s

orchestration_failure · router_soft · regulator_runner/T1090-W1FSA/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-05

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.; receipt_id backfilled by tools/error_ledger/triage.py: TR-118296 score=75 reason=error_code+chain_step+domain+time<=43158s

orchestration_failure · router_soft · regulator_runner/T1090-W1RPST/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-05

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.; receipt_id backfilled by tools/error_ledger/triage.py: TR-118296 score=75 reason=error_code+chain_step+domain+time<=47696s

orchestration_failure · router_soft · regulator_runner/T1090-W1ODM/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-05

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.; receipt_id backfilled by tools/error_ledger/triage.py: TR-118296 score=75 reason=error_code+chain_step+domain+time<=49191s

orchestration_failure · router_soft · regulator_runner/RS-10/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-05

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.; receipt_id backfilled by tools/error_ledger/triage.py: TR-138047 score=80 reason=task_id+chain_step+time<=2361s

orchestration_failure · router_soft · regulator_runner/T1090-W2XCUT/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-06

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T1277/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-06

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T1386/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-06

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T1386/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-06

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T1345/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-08

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T3205/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-08

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T3204/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-08

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T1400/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-08

regulator_runner main_environment_required: regulator must run in the main environment, not a Codex sandbox

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T1400/main_environment_required

tools/infra_core/regulator_runner.py:run_regulator

复发簇 · auto_regulator_runner:codex_shell_requires_explicit_allow:no_tier:codex_shell_requires_explicit_allow

10 次
错误台账unresolved2026-07-08

regulator_runner codex_shell_requires_explicit_allow: regulator default dispatch from Codex shell requires explicit REGULATOR_ALLOW_CODEX_SHELL=1 plus TODO_ID and CODEX_THREAD_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T1400/codex_shell_requires_explicit_allow

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-08

regulator_runner codex_shell_requires_explicit_allow: regulator default dispatch from Codex shell requires explicit REGULATOR_ALLOW_CODEX_SHELL=1 plus TODO_ID and CODEX_THREAD_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T3224/codex_shell_requires_explicit_allow

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-08

regulator_runner codex_shell_requires_explicit_allow: regulator default dispatch from Codex shell requires explicit REGULATOR_ALLOW_CODEX_SHELL=1 plus TODO_ID and CODEX_THREAD_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T1400/codex_shell_requires_explicit_allow

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-08

regulator_runner codex_shell_requires_explicit_allow: regulator default dispatch from Codex shell requires explicit REGULATOR_ALLOW_CODEX_SHELL=1 plus TODO_ID and CODEX_THREAD_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T1400/codex_shell_requires_explicit_allow

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-10

regulator_runner codex_shell_requires_explicit_allow: regulator default dispatch from Codex shell requires explicit REGULATOR_ALLOW_CODEX_SHELL=1 plus TODO_ID and CODEX_THREAD_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/MIG-REB-P2/codex_shell_requires_explicit_allow

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-10

regulator_runner codex_shell_requires_explicit_allow: regulator default dispatch from Codex shell requires explicit REGULATOR_ALLOW_CODEX_SHELL=1 plus TODO_ID and CODEX_THREAD_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/MIG-REB-P2/codex_shell_requires_explicit_allow

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-11

regulator_runner codex_shell_requires_explicit_allow: regulator default dispatch from Codex shell requires explicit REGULATOR_ALLOW_CODEX_SHELL=1 plus TODO_ID and CODEX_THREAD_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/AUD-P3-FORENSIC-READ/codex_shell_requires_explicit_allow

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-12

regulator_runner codex_shell_requires_explicit_allow: regulator default dispatch from Codex shell requires explicit REGULATOR_ALLOW_CODEX_SHELL=1 plus TODO_ID and CODEX_THREAD_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/DDP-CCTX-R007/codex_shell_requires_explicit_allow

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-16

regulator_runner codex_shell_requires_explicit_allow: regulator default dispatch from Codex shell requires explicit REGULATOR_ALLOW_CODEX_SHELL=1 plus TODO_ID and CODEX_THREAD_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T2336/codex_shell_requires_explicit_allow

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-18

regulator_runner codex_shell_requires_explicit_allow: regulator default dispatch from Codex shell requires explicit REGULATOR_ALLOW_CODEX_SHELL=1 plus TODO_ID and CODEX_THREAD_ID

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

orchestration_failure · router_soft · regulator_runner/T3211/codex_shell_requires_explicit_allow

tools/infra_core/regulator_runner.py:run_regulator

复发簇 · opencode_completion_detection

9 次
错误台账workaround2026-07-01

deepseek:high dispatch 返回 fake_success: turns=0 cost=0 tokens=0 duration=142s, result_path=null, 未产出 eval verdict. router 检测到 fake_success 但仍标 dispatch OK 且未 fallback

展开

处置:先同通道重试一次验证是否瞬时; 若复现则 deepseek(OpenCode 通道) eval 暂不可用, 升级用户(需修 opencode completion detection); S0 eval 以 2/3 (Codex+Opus 双 PASS) 继续, 不让 fallback 冒充 deepseek

provider_silent_failure · router_soft · Run1/S0/eval/attempt1

/tmp/session_dispatch_runs/s0_eval_deepseek_S0_eval_deepseek_02ac71b3/router.log:4

错误台账unresolved2026-07-01

deepseek:high 重试仍 fake_success (turns=0, result_path=null). 确认非瞬时, deepseek(OpenCode 通道) eval 输出不可用

展开

处置:确认 deepseek eval 通道 fake_success 复现. 升级用户三选一: (a) 修 opencode completion detection 让 deepseek 真产出; (b) 换第三评价者(claude-zai/kimi); (c) 暂以 Codex+Opus 双评价者. S0 eval 最终 = 2/3 (Codex+Opus 双 PASS)

provider_silent_failure · router_soft · Run1/S0/eval/attempt2_retry

/tmp/session_dispatch_runs/s0_eval_deepseek_retry_S0_eval_deepseek_retry_*/router.log; 关联 ERR-000001

错误台账workaround2026-07-01

根因定位: deepseek-v4-pro 是 reasoning 模型, 答案进 reasoning_content, 正式 content 返回空(finish_reason=length); opencode 通道只读 content -> 0-turn -> router 判 fake_success. 非 auth/模型名问题(API 直连验证: key 有效, /v1/models 含 v4-pro)

展开

处置:三选一: (a)opencode deepseek 适配器读 reasoning_content+调大 max_tokens; (b)改用 deepseek-v4-flash(疑非reasoning); (c)第三评价者用 GLM claude-zai:high(已验证真跑 turns=11 cost0.64 modelUsage=glm-5.2[1m]) 顶替. 已采用(c)默认, (a)deepseek修复列backlog

provider_silent_failure · router_soft · Run1/S0/eval/rootcause

curl https://api.deepseek.com/v1/chat/completions model=deepseek-v4-pro -> content='' reasoning_content非空 finish_reason=length

错误台账workaround2026-07-06

replay_runner.py --entry cc --channel ollama:high hangs to 200s timeout (exit 124); worker_prompt.replay.txt written but no completion/dispatch_meta emitted. Confirms retest-inventory底表 NO_DATA #2: ollama:high routing v…

展开

处置:B-path真跑改用: (a) required_channel any 能力优先 codex:high 或原生 claude:high (已实证通道); (b) 若要 GLM/Kimi via Ollama, 走 claude-ollama-glm CC wrapper 直调而非 replay_runner --channel ollama:high; (c) ollama:high 经 replay_runner 的挂死单列修复票待查 router.py dispatch_helper 解析

tool_failure · silent · TSM-r3 B-path channel probe: replay_runner --entry cc --channel ollama:high

/private/tmp/claude-501/-Users-lexuanzhang-context-infra/9c492bb8-6aad-416d-89cb-477f547f61b3/tasks/bn4ky36fb.output

错误台账resolved2026-07-06

RESOLVED: ollama:high 挂死根因=router TIER_CONFIG channel=opencode(经 OpenCode server,headless 不返回)。eval_harness CC-entry 正确通道=claude-ollama-glm:high(CC wrapper 直调 claude 二进制打 Ollama Cloud Anthropic-compat 端点)。真探针 replay_run…

展开

处置:replay_runner 加 --model-override 透传(dispatch_helper 已支持,Kimi-via-Ollama 用 --model-override kimi-k2.6)+ docstring/CLI help 明确 GLM/Kimi via Ollama 走 claude-ollama-glm:high 非 ollama:high。B 路语义真跑通道解锁。

tool_failure · silent · TSM-r3 TSM-OLLAMA-DISPATCH resolution

tools/eval_harness/replay_runner.py:run_cc_replay

错误台账resolved2026-07-20

resolution update for ERR-000001: 行为已在现行 router.py 修复:degenerate success(含 fake_success/turns=0)在 tier 内标记失败并 continue 到下一 tier(router.py:1520-1539,fallback 已接,非当年'检测到但仍标OK不fallback');OpenCode 通道另于 :2266-2273 加 syntheti…

展开

处置:行为已在现行 router.py 修复:degenerate success(含 fake_success/turns=0)在 tier 内标记失败并 continue 到下一 tier(router.py:1520-1539,fallback 已接,非当年'检测到但仍标OK不fallback');OpenCode 通道另于 :2266-2273 加 synthetic turns=1 修复 0-turn 全量误判。另见 ERR-000091(ollama:high 挂死根因定位+通道纠正)与 07-18 provider chain 定版(deepseek 评价者已被 GLM 顶替,不在现行 11-tier 链内)。谓词命中的是陈旧未回填状态位,非活体问题。

provider_silent_failure · router_soft · Run1/S0/eval/attempt1

tools/cli_agent/router.py:1520-1539,tools/cli_agent/router.py:2266-2273

错误台账resolved2026-07-20

resolution update for ERR-000002: 同簇 ERR-000001 修复覆盖:degenerate success 在 router.py:1520-1539 已改为 tier 内失败+continue fallback,OpenCode 通道 :2266-2273 加 synthetic turns=1 消除 0-turn 误判。deepseek 评价者已被 GLM 顶替,不在 07-18 定版 11-t…

展开

处置:同簇 ERR-000001 修复覆盖:degenerate success 在 router.py:1520-1539 已改为 tier 内失败+continue fallback,OpenCode 通道 :2266-2273 加 synthetic turns=1 消除 0-turn 误判。deepseek 评价者已被 GLM 顶替,不在 07-18 定版 11-tier 链内,本簇不再复发。

provider_silent_failure · router_soft · Run1/S0/eval/attempt2_retry

tools/cli_agent/router.py:1520-1539,tools/cli_agent/router.py:2266-2273

错误台账resolved2026-07-20

resolution update for ERR-000003: 同簇 ERR-000001 修复覆盖:degenerate success 在 router.py:1520-1539 已改为 tier 内失败+continue fallback,OpenCode 通道 :2266-2273 加 synthetic turns=1 消除 0-turn 误判。deepseek-v4-pro reasoning_content 空 co…

展开

处置:同簇 ERR-000001 修复覆盖:degenerate success 在 router.py:1520-1539 已改为 tier 内失败+continue fallback,OpenCode 通道 :2266-2273 加 synthetic turns=1 消除 0-turn 误判。deepseek-v4-pro reasoning_content 空 content 问题连带 deepseek 评价者被 GLM 顶替(07-18 定版)而消解,不在现行 11-tier 链内。

provider_silent_failure · router_soft · Run1/S0/eval/rootcause

tools/cli_agent/router.py:1520-1539,tools/cli_agent/router.py:2266-2273

错误台账resolved2026-07-20

resolution update for ERR-000090: 根因已定位并修复:ollama:high 挂死根因=router TIER_CONFIG channel=opencode(经 OpenCode server headless 不返回);正确通道为 claude-ollama-glm:high(见 ERR-000091 完整修复记录:replay_runner --model-override 透传落地 + docs…

展开

处置:根因已定位并修复:ollama:high 挂死根因=router TIER_CONFIG channel=opencode(经 OpenCode server headless 不返回);正确通道为 claude-ollama-glm:high(见 ERR-000091 完整修复记录:replay_runner --model-override 透传落地 + docstring/CLI help 明确通道纠正,真探针实测 exit 0/turn_completed)。

tool_failure · silent · TSM-r3 B-path channel probe: replay_runner --entry cc --channel ollama:high

ERR-000091

复发簇 · tool_failure

8 次
错误台账resolved2026-07-01

solid_decision_phase_gate run_regulator 成功路径 TypeError:_sd4_status() 旧签名二参调用(缺 expected_linkage_ids/expected_resolved_question_ids)。长期被 regulator_runner 的 missing-TODO_ID 早退掩盖(从未走到成功路径=静默盲区),首次带 TODO_ID 拿到 codex 回包即崩

展开

处置:check.py run_regulator 改四参调用(传已算好的 expected_linkage/expected_backfill);复现命令:TODO_ID=<id> python3 tools/solid_decision_phase_gate/check.py --asset <md> --regulator-tier codex:high; receipt_id backfilled by tools/error_ledger/triage.py: TR-057467 score=70 reason=task_id+time<=351s

tool_failure · silent ·

tmp/diagram_learning_systems_run_20260701/sd_gate_full2.json traceback(check.py L476)

错误台账resolved2026-07-04

W0-B 修复验证:BUG-R5-07 wait marker Stop feedback 自唤醒环 + BUG-R5-09 receipt cutoff updated_ts 截断假阴性;隔离 pytest 75 passed,wait marker 3 events/1 feedback,cutoff passed=True

展开

处置:Added wait-marker feedback throttle with preserved phase_waiting events; added cursor phase_entered_ts and receipt cutoff fallback; verified in isolated env.; receipt_id backfilled by tools/error_ledger/triage.py: TR-121457 score=80 reason=task_id+chain_step+time<=3441s

tool_failure · silent · Run5/W0-B

adhoc_jobs/context_infra_base_tooling_buildout_20260615/Run5/run_artifacts/W0-B/RESULT.md

错误台账resolved2026-07-04

W0-D1 闭环:RR-4 自动 prompt 捕获 live(每轮逐字落库+域建议+intake 候选标记)

展开

处置:landed; kill-switch RR4_CAPTURE_DISABLE

tool_failure · silent · run5/W0-D1

commit 63a791950; contexts/prompt_records/capture_2026-07-04.jsonl 12条

错误台账resolved2026-07-04

W0-E 闭环:gate --task→TODO_ID 传导+记分牌仅缺口行生成 intake 接头

展开

处置:landed

tool_failure · router_soft · run5/W0-E

commit 63a791950; ERR-000011 对应缺口

错误台账workaround2026-07-04

Stop hook batch git add fails silently: stderr swallowed by --pathspec-from-file --ignore-errors; aborted/failed commits leave ~16.7 GiB unreachable loose objects and linearly accumulating session indexes / sharedindex …

展开

处置:Deployed diagnostic logging (_diag_log/_log_batch_failure in commit_wrapper.sh) wired via GIT_COMMIT_WRAPPER_DIAG_LOG from stop_hook.sh; future failures will capture stderr + rc + sample paths. Heavy-load skip switches documented. Orphan/session-index cleanup deferred to proposal (prune / session-death policy).

tool_failure · silent · run5/W0-C/batch-add

adhoc_jobs/context_infra_base_tooling_buildout_20260615/Run5/run_artifacts/W0-C/RESULT.md

错误台账workaround2026-07-04

BUG-R5-01 root cause FIXED (not just diagnostics): batched git add was deterministically fatal because commit_wrapper passed an extra :(exclude) pathspec arg alongside --pathspec-from-file, which git rejects ('--pathspe…

展开

处置:Root cause fixed at commit_wrapper.sh:578-584 (also session_merge.sh). Diagnostics (_diag_log/_log_batch_failure via GIT_COMMIT_WRAPPER_DIAG_LOG) + heavy-load switches (GIT_STOP_HOOK_SKIP_SHADOW_COMPARE_OVER_FILES, GIT_STOP_HOOK_MAX_FILES_PER_COMMIT, default 0) landed. Session-index/sharedindex reaping ALREADY exists+scheduled (orphan_session_index_reaper.py, crontab 20 3 * * *). Remaining: loose-object prune (git prune --expire=2.weeks.ago in daily_digest) deferred to RESIDUE_GOVERNANCE_PROPOSAL.md await_user; rc=124 300s serialize timeout is an independent path listed as residual.

tool_failure · silent · run5/W0-C/batch-add-rootcause

adhoc_jobs/context_infra_base_tooling_buildout_20260615/Run5/run_artifacts/W0-C/RESULT.md; diagnostics/w0c_acceptance_verify.py (root-cause pre/post repro + orphan sim + deploy safety, all PASS); tools/git/tests/test_w0c_batch_add.py::GitlinkExcludeBatchRegressionTests; live stop_hook.log 'batched git add ok: paths=76'

错误台账unresolved2026-07-21

infra 站共享消毒器泄漏 tilde 路径:build.mjs FORBIDDEN 黑名单与 decisions_data/dashboard_data 的 clean() 只脱敏 /Users/<user> 绝对路径和一个字面 ⟨CC 记忆目录⟩ 前缀,通用 ~/ 路径全漏。线上实测 dist/decisions.html 33 处未脱敏(~/.cache/context-infra、~/.claude.json、~/.code…

展开

处置:建议:把 build_pending_board.mjs 的宽 tilde 正则移植进共享 clean()/FORBIDDEN 并全站回归(26 页隐私断言重跑);作为独立小改动走,勿与功能变更混批

tool_failure · silent ·

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/site_candidates/kimi/dist/decisions.html; 发现者=board-port worker 2026-07-21; 宽 tilde 正则对照实现在同目录 tools/build_pending_board.mjs

错误台账workaround2026-07-21

wrangler pages deploy 输出吞没复发(r002 已知坑加剧版):不仅管道下,文件重定向下最终 deployment URL 行也会丢(本次三跑:一次假成功未产 deployment、一次只出 banner、一次 exit=0 停在 Deploying... 但服务端实际成功)。输出解析不可作为部署成败依据

展开

处置:已定加固验证法:部署成败一律以 wrangler pages deployment list 的新 Production 行(source commit 匹配)+ 部署专属 URL 内容 grep(记得 -L 跟 308)为准,永不解析 deploy 命令输出

tool_failure · silent ·

session 79195072 scratchpad/wr_deploy.log(exit=0 无 URL 而 deployment 33e0003e 实际创建成功);对照 STATE.md 2026-07-19 09:5x 原案

复发簇 · wrangler_false_success

4 次
教训2026-07-19

部署假成功:wrangler 重定向到文件时 rc=0 但只输出 banner、实际未部署

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r002_site_overhaul_20260719/reports/FINAL_REPORT.md

教训2026-07-20

部署纪律复验:wrangler 裸跑+仓根路径+rc+http_code+逐字节对 dist,缺一即假成功温床

展开

根因:

对策:wrangler 裸跑+仓根路径+rc+http_code+逐字节对 dist 五件套

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r003_site_remediation_20260720/reports/FINAL_REPORT.md

教训2026-07-21

wrangler 假成功新形态坐实:stdin 为 socket 时横幅后静退 rc=0

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r005_dashboard_redesign_20260721/reports/FINAL_REPORT.md

教训2026-07-19

wrangler 输出在管道/命令替换下静默、ANSI 色码会骗过 grep ERROR、Pages 对 *.html 路径 308 重定向

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/synthesis/FINAL_REPORT.md

复发簇 · codex_managed_profile_git_index_lock_denied

4 次
错误台账unresolved2026-07-06

commit_wrapper could not stage requested S3 gate paths because the managed Codex filesystem profile denied creation of .git/index.lock; HEAD commit could not be produced from this session.

展开

处置:No workaround available inside current approval policy/sandbox; continue writing workspace artifacts and report commit blockage explicitly.

environment · http_hard · r024_w5h_dc_s3/segment_commit_1

commit_wrapper stderr: fatal: Unable to create .git/index.lock: Operation not permitted

错误台账deferred2026-07-06

W5M scoped git add for scheduling bridge files failed because current Codex sandbox cannot create .git/index.lock: Operation not permitted.

展开

处置:All implementation, tests, reports, standing queue, todo completion, and resolve replay are on disk; commit must be rerun from a session with .git write permission.

environment · http_hard · r030_w5m_scheduling_bridge/git-stage-commit

terminal: git add -> fatal: Unable to create '.git/index.lock': Operation not permitted

错误台账deferred2026-07-06

W5-L1 scoped git add/commit failed because current Codex sandbox cannot create .git/index.lock: Operation not permitted.

展开

处置:All implementation, tests, todo updates, error ledger rows, and report artifacts are on disk; commit must be rerun from a session with .git write permission using the whitelist in REPORT.md/final response.

environment · http_hard · r029_w5l1_hook_capture/git_add_commit

terminal: git add ... -> fatal: Unable to create '.git/index.lock': Operation not permitted

错误台账deferred2026-07-06

W5P scoped git add failed: fatal: Unable to create .git/index.lock: Operation not permitted

展开

处置:All scoped files are on disk; commit must be run from an environment with .git/index.lock write permission using the whitelist in REPORT.md/final response.

environment · http_hard · W5P/git_add_commit

adhoc_jobs/systems_consolidation_20260705/runs/r033_w5p_git_nightwindow/REPORT.md

复发簇 · stop_hook_dpath_stale_baseline_mass_delete_rc126_orphan

4 次
错误台账resolved2026-07-18

陈旧 session(自07-10挂起) turn 41 Stop hook 把1407条相对陈旧baseline看似被删的路径git rm --cached进主索引(session index env缺失回落主索引),随后commit_wrapper.sh rc=126崩溃、无可归因stderr,1407条staged删除孤儿化(事后核实1358幽灵+49真删);根因与ERR-000138同源(commit链上组件同样撞Python…

展开

处置:main_index_reconcile.sh 人工跑一轮清理:ok=1358 fail=0,staged D 1407→49(全为合法在途真删);结构修复=stop_hook.sh/commit_wrapper.sh 解释器改绑GIT_TOOLING_PYTHON3(同ERR-000138)+D-path新增GIT_STOP_HOOK_MAX_DELETED_PATHS=200幅度护栏(超阈值本turn不提交删除路径,changed paths仍正常提交)+rc=126/127时补充exec-failure诊断日志行(wrapper可执行性/解释器版本/gateway状态/PATH)弥补本次取证盲区

orchestration_failure · silent · stop_hook/commit_wrapper/session-ref-only

stop_hook.log:38456-38469 (rc=126 无 stderr); .git/index mtime 01:07; contexts/survey_sessions/git_tracking_anomaly_forensics_20260718_manual.md §1 事件A + §3 当前状态核验表

错误台账workaround2026-07-18

resolution_status 更正(ERR-000139 修正): 团队复核后确认 rc=126 的精确崩溃机理未坐实——已排除 gateway serialize import 链/die()全部路径/pre-commit hook 直接导致;时序线索指向 orphan-guard 预算撞线(stop_hook.log 23:07:19 orphan guard armed budget 42s of 45s elapsed=…

展开

处置:结构性修复已落地(解释器改绑+D-path护栏+rc=126/127诊断埋点),但rc=126本身的确切触发机理待下次真实复现坐实;观察窗判据=埋点上线后30天无新rc=126或出现一次即按诊断快照定谳

orchestration_failure · silent · stop_hook/commit_wrapper/session-ref-only

contexts/survey_sessions/git_tracking_anomaly_forensics_20260718_manual.md; adhoc_jobs/context_infra_migration_20260709/design/OPEN_PROBLEMS.md GIT-4; ~/.cache/context-infra/stop_hook.log 2026-07-17T23:07:19Z/23:07:56Z

错误台账resolved2026-07-18

修正 ERR-000139(及其修正记录 ERR-000141)对 rc=126 的根因归因: 解释器换绑并未解决 rc=126, 本条为真实根因。2026-07-18 晚间复现定谳: commit_wrapper.sh 的 session_stage_allowlist_dirs() 把 session index 相对 HEAD 的 adhoc_jobs 全量 diff 路径清单通过环境变量传给 python 子进程; 该清单随 …

展开

处置:三件套修复: (1) session_stage_allowlist_dirs() 路径清单改走 mktemp 临时文件传递, 消灭环境变量的 ARG_MAX 天花板; (2) 该辅助staging步骤失败改为非致命(WARN到stderr不再吞掉+跳过allowlist补充步骤+主commit继续); (3) stop_hook.sh SYSMSG发射器加 errors=replace 加固, 修复其自身对surrogate字符的UnicodeEncodeError崩溃。复现方法=一次性probe ref+bash -x全链路trace, 先复现RC=126, 修复后同probe场景RC=0且真实落commit(hash 6614c502)。回归=guard套件58/58与3.9钉死17/17保持全绿; invariant_stop_hook.sh 独立单次运行两次均24/24 PASS, 但与其他重负载测试连续运行时观测到一次单案例瞬时FAIL, 疑似并发资源竞争非本次修复引入的稳定回归, 建议后续以单独运行结果为准。

tool_failure · silent · stop_hook/commit_wrapper/session_stage_allowlist_dirs

tools/git/commit_wrapper.sh:334-347(BUG-021 root cause 2026-07-18 代码注释,逐字记录ARG_MAX/E2BIG机理); tools/git/commit_wrapper.sh:395(WARN非致命化实现); tools/git/stop_hook.sh:55-56(SYSMSG surrogate加固注释); 一次性 probe ref refs/sessions/rc126probe(已删除,提交对象仍在object db, git cat-file -t 6614c502 返回commit已核验存在); tools/git/BUG_TRACKER.md BUG-021 2026-07-18晚补充定谳段

错误台账resolved2026-07-18

登记结构收口: rc=126 的 E2BIG 根因(session_stage_allowlist_dirs 环境变量传参 1,101,473 bytes 越过 macOS ARG_MAX, execve 返回 argument list too long, bash 表现为退出码126)已在 ERR-000143 完整记录(逐字机理/probe复现/三件套修复/回归数据, 本条不重复展开)。ERR-000139 的 effectiv…

展开

处置:无新增技术修复(技术修复已在 ERR-000143 记录并落地); 本条是 BUG_TRACKER 从单条 BUG-021 拆分为 BUG-021(解释器墙)+BUG-023(ARG_MAX/E2BIG)两个独立条目后, error_ledger 侧的对应登记结构收口, 确保三个登记簿(BUG_TRACKER/OPEN_PROBLEMS/error_ledger)对同一事实的编号口径彼此一致可查

tool_failure · silent · stop_hook/commit_wrapper/session_stage_allowlist_dirs

contexts/runtime/error_ledger/errors.jsonl ERR-000139/ERR-000141/ERR-000143 三条记录链; tools/git/BUG_TRACKER.md BUG-023(#bug-023, 新增独立条目) 与 BUG-021 状态回收为 FIXED; adhoc_jobs/context_infra_migration_20260709/design/OPEN_PROBLEMS.md GIT-4 残差①CLOSED

复发簇 · environment

3 次
错误台账workaround2026-07-04

phase 驱动器经 gate_registry 裸跑 check_report_completeness 时无 TODO_ID env,regulator_runner 以 missing_task_id 早退,表层报 no_json_object_found(与 round4 requirement_analysis_gate 同款);registry 120s 超时也不足以跑 codex:high regulator

展开

处置:带 TODO_ID 手动跑全 profile 留证,plan 谓词改锚 artifact_contains 标记;根治候选=registry 传递 task_id 或 gate 侧 TODO 缺省降级 shell-only

environment · silent ·

adhoc_jobs/dark_context_system_20260704/plan.yaml P7 exit 注释 + gate_run/report_gate_result.json(带 TODO_ID 手动跑 final PASS 对照)

错误台账deferred2026-07-04

Disk hit 100% (4.1G free of 460G) at dark_context round2 start; transient 'no space left on device' during shell output capture. .git regrown to 19G: objects 12G + session-indexes 2.7G (447 files) + 148 sharedindex ~4.1…

展开

处置:none yet; surfaced to user in round2 report; workers instructed disk-frugal

environment · silent ·

.git du output in session a7f1a122; reaper dry-run scanned=296 orphan=8 stale=0

错误台账unresolved2026-07-21

infra 站每日 publish cron(08:15)静默失败:publish.mjs 子进程 spawn npx/node 依赖 PATH,cron 环境无 nvm PATH,日志唯一输出=env: node: No such file or directory。crontab 行本身用了绝对 node 路径但子进程未继承等效 PATH。站点自动发布事实停摆(最新成功发布均为 session 手动)

展开

处置:建议 publish.mjs 内 spawn 用 process.execPath 与其同目录 npx 的绝对路径(或 env 注入 PATH),修后看次日 site_publish.log 首行

environment · silent ·

~/.cache/context-infra/site_publish.log tail;crontab 行=15 8 * * * …/node tools/publish.mjs --if-changed;D-SYN-6 同族(cron PATH/解释器)

复发簇 · regulator_todo_id_env

3 次
错误台账workaround2026-07-04

requirement_analysis_gate 的 --task 参数不传导为 regulator 的 TODO_ID 要求,裸跑必 FLAG missing_task_id;错误信息未告知需设 TODO_ID 环境变量(可操作性缺口,同签名 ERR-000007/ERR-000009 第三次复现)

展开

处置:TODO_ID=T1090 环境变量 workaround;根治=--task 映射 regulator env 或报错给修法,挂 BUG-R5-10 入 Run5 W0; receipt_id backfilled by tools/error_ledger/triage.py: TR-121457 score=90 reason=task_id+chain_step+time<=430s

tool_failure · router_soft · run5/P2/requirement_analysis_gate

gate_result.json runner.reason=missing_task_id (with --task BT-RUN5-001-P2); TODO_ID=T1090 后 regulator 真跑 RAW_DELIVERED

错误台账resolved2026-07-04

BUG-R5-10 resolved: requirement_analysis_gate --task propagates to regulator TODO_ID; missing_task_id error message now gives actionable export TODO_ID=<task-id> or --task <task-id> command

展开

处置:--task mapped through check.py → regulator_adapter → os.environ[TODO_ID]; regulator_runner missing_task_id required_next_action updated to actionable command; BUG-R5-08 scoreboard covered rows no longer emit todo intake

tool_failure · router_soft · run5/P2/requirement_analysis_gate

Resolves ERR-000011; adhoc_jobs/context_infra_base_tooling_buildout_20260615/Run5/run_artifacts/W0-E/RESULT.md; adhoc_jobs/context_infra_base_tooling_buildout_20260615/Run5/run_artifacts/W0-E/test_output.log

错误台账resolved2026-07-04

ERR-000011 闭环 (BUG-R5-10): requirement_analysis_gate --task 现端到端传导为 regulator TODO_ID(regulator_adapter.run_raw_json_regulator 在调用 run_regulator 前设 TODO_ID=task.upper() 并在 finally 还原);missing_task_id 时默认 human 输出新增 regu…

展开

处置:已验证:run_raw_json_regulator(task='t1090')→run_regulator 内 TODO_ID=T1090、调用后还原 None;missing_task_id payload→runner.reason+required_next_action('export TODO_ID=<task-id> or pass --task <task-id>');_human 现输出 regulator_fix。回归:requirement_analysis_gate 15 + test_health 4 + filter_compress_gate 18 + semantic_decision_gate 3 + deferred_verification_gate 11 + session_claim_audit_gate 17 全绿。option1/structured 侧由 commit 63a7919509 落地,本 session 补 human-mode surfacing 残差。

tool_failure · router_soft · run5/W0-E/requirement_analysis_gate

tools/requirement_analysis_gate/check.py:_human_regulator; tools/infra_core/regulator_adapter.py:run_raw_json_regulator; tools/test_health/scoreboard.py:_has_gap_cells; adhoc_jobs/context_infra_base_tooling_buildout_20260615/Run5/run_artifacts/W0-E/RESULT.md

复发簇 · orchestration_failure

3 次
错误台账workaround2026-07-04

Background bash tasks (codex exec workers) killed in simultaneous pairs twice (~21:50 and ~22:25) during dark_context round2; zero output flushed; orphan codex_bridge procs left. Pattern matches Run3 stall crisis; relau…

展开

处置:tmux new-session detached for s1cont/s2cont with log+done markers

orchestration_failure · silent ·

tasks bb8tvlrvs/b968fm750/b0g8f3o3b/bxzrjzaxr all status=killed, 0-byte outputs

错误台账resolved2026-07-05

W3 队列 task_ids 凭记忆猜票号(T1152/T1156 实为 QGF/PRS 无关票),dispatcher 据此 start 错票、检查者 done 错票后自查发现;两票已经合法状态机回滚 pending

展开

处置:回滚两票;W3 收口证据锚 runs/r010-r012 产物非票号;纪律=队列 task_ids 必须 todo show 核题后填写禁凭记忆

orchestration_failure · silent · w3_queue_authoring

dispatch_queue_w2.yaml W3A/W3B 条目 + dispatch_receipt_w1.md 勘误段

错误台账resolved2026-07-06

W4 queue entry task anchor wrong: task_ids=[T1240] but T1240 is SES-5 F04 guardian-daemon fix ticket. Third instance of queue-anchor-by-memory family (prior: T1152/W3A, T1156/W3B)

展开

处置:Registered dedicated ticket T1279 and corrected queue task_ids before any ignition; comment left in YAML

orchestration_failure · silent ·

design_prep/dispatch_queue_w2.yaml W4-INTEGRAL-ACCEPTANCE; todo show T1240 title=SES-5 F04: land guardian daemon default trigger path

复发簇 · git_index_lock_eperm

3 次
错误台账unresolved2026-07-05

commit_wrapper could not stage Q15 paths because git add failed to create .git/index.lock: Operation not permitted in current Codex sandbox

展开

处置:Current session cannot write .git; keep Q15 files in worktree, record blocker, and require unrestricted/TCC-authorized session to rerun the same commit_wrapper command

environment · http_hard · r004/P2/q15_commit_attempt

错误台账unresolved2026-07-06

T1277 scoped git add for r020 PRS tail artifacts failed because the managed Codex sandbox cannot create .git/index.lock: Operation not permitted.

展开

处置:Artifacts, RCL rows, and todo completed state remain in the worktree; rerun scoped git add/commit from an unrestricted main environment if a commit is required.

environment · http_hard · T1277/scoped_git_add

terminal: fatal: Unable to create '.git/index.lock': Operation not permitted

错误台账unresolved2026-07-06

W5-B commit_wrapper could not stage or commit verified T1249/T1250/T1251/T1252/T1253/T1248 artifacts because current Codex sandbox cannot create .git/index.lock: Operation not permitted.

展开

处置:Implementation, tests, todo completion events, and report remain in worktree; rerun the same scoped commit_wrapper command from an unrestricted main environment to land in HEAD.

environment · http_hard · r017_w5b_ledger_liveness/commit_wrapper

terminal: commit_wrapper.sh fatal: Unable to create '.git/index.lock': Operation not permitted; report=adhoc_jobs/systems_consolidation_20260705/runs/r017_w5b_ledger_liveness/REPORT.md

复发簇 · provider_behavior

3 次
错误台账unresolved2026-07-06

T1153 canary from MAIN env: codex-oc:high returns degenerate_success (empty/degenerate reply via OpenCode gpt-5.5); sandbox network was not the only blocker. Channel currently not viable as true-hit tier

展开

处置:T1153 verify-failed back to in_progress; unlock predicate=non-degenerate reply recorded to coverage ledger; feeds T1273 tier-viability config

provider_behavior · router_soft ·

router --verbose 02:1x: tier=codex-oc:high failed kind=degenerate_success; fallback kimi=quota-dead, opencode:high timed out at 90s wrapper

错误台账deferred2026-07-06

W5C worker died silently at 61 turns/22min: is_error=true with empty result and no api_error_status (glm channel)

展开

处置:channel demoted for the night; investigation ticket T1475

provider_behavior · silent ·

/tmp/dispatch_edge_runs/W5C-REQ-BRIDGE-TAIL/2026-07-06T011941_0200/session_dispatch/W5C-REQ-BRIDGE-TAIL_result.json

错误台账deferred2026-07-06

claude-zai channel degraded ~07:55 after heavy overnight use: rc=0 with EMPTY output (degenerate), router classifies subprocess_error and falls back kimi(quota-dead)->opencode. W5O2/W5G2-RESUME2 died on this

展开

处置:remaining lanes shifted to codex:high; zai recovery probe deferred to next coordinator session; folds into T1475/T1273 tier-viability

provider_behavior · silent ·

direct probe 08:15 'claude-zai -p' rc=0 zero reply text; router.log W5O2 tier=claude-zai:high failed kind=subprocess_error

复发簇 · auto_unified_gate:main_environment_required:codex:high

3 次
错误台账unresolved2026-07-06

unified_gate regulator failure main_environment_required: main_environment_required

展开

处置:Inspect regulator runner output and gate parser contract.

tool_failure · router_soft · unified_gate/regulator/main_environment_required

tools/unified_gate/base.py:Regulator.run

错误台账unresolved2026-07-06

unified_gate regulator failure main_environment_required: main_environment_required

展开

处置:Inspect regulator runner output and gate parser contract.

tool_failure · router_soft · unified_gate/regulator/main_environment_required

tools/unified_gate/base.py:Regulator.run

错误台账unresolved2026-07-08

unified_gate regulator failure main_environment_required: main_environment_required

展开

处置:Inspect regulator runner output and gate parser contract.

tool_failure · router_soft · unified_gate/regulator/main_environment_required

tools/unified_gate/base.py:Regulator.run

复发簇 · auto_unified_gate:missing_task_id:codex:high

3 次
错误台账unresolved2026-07-11

unified_gate regulator failure missing_task_id: missing_task_id

展开

处置:Inspect regulator runner output and gate parser contract.

tool_failure · router_soft · unified_gate/regulator/missing_task_id

tools/unified_gate/base.py:Regulator.run

错误台账unresolved2026-07-11

unified_gate regulator failure missing_task_id: missing_task_id

展开

处置:Inspect regulator runner output and gate parser contract.

tool_failure · router_soft · unified_gate/regulator/missing_task_id

tools/unified_gate/base.py:Regulator.run

错误台账unresolved2026-07-19

unified_gate regulator failure missing_task_id: missing_task_id

展开

处置:Inspect regulator runner output and gate parser contract.

tool_failure · router_soft · unified_gate/regulator/missing_task_id

tools/unified_gate/base.py:Regulator.run

复发簇 · git_restore_overreach

2 次
教训2026-07-19

已入库文件走 gateway 反向 commit 或 git revert——禁 stash/reset/clean(GIT-3/INC-1 教训)

展开

根因:

对策:已入库文件回滚只走 gateway 反向 commit 或 git revert,禁 stash/reset/clean

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/design/D2_FINAL.md

教训2026-07-20

INCIDENT-r003-1:worker「恢复现场」越界把 build.mjs/svg.mjs/cleanctx_page.mjs 回滚到 HEAD,毁掉三 worker 未提交改动

展开

根因:

对策:派发 charter 必写:禁一切 git 恢复类命令(checkout/restore/stash/reset/clean)

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r003_site_remediation_20260720/reports/FINAL_REPORT.md

复发簇 · tmux_survives_session_cleanup

2 次
教训2026-07-20

tmux 承载长任务:session 后台任务被清理 12 次,tmux 内 worker 全部存活

展开

根因:

对策:长任务走 tmux 承载

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r003_site_remediation_20260720/reports/FINAL_REPORT.md

教训2026-07-05

两轮 harness 后台批量收割(恢复=tmux+续做 worker+Monitor 巡检)

展开

根因:

对策:恢复=tmux+续做 worker+Monitor 巡检

adhoc_jobs/requirement_decomposition_context_isolation_20260704/reports/FINAL_REPORT_R002_20260704.md

复发簇 · concurrent_pollution_false_positive

2 次
教训2026-07-11

并发污染假阳:eval run 与实装 worker 并行导致 isolation 误报

展开

根因:

对策:eval 与实装不并行(教训已入 ledger)

adhoc_jobs/context_infra_base_tooling_buildout_20260615/methodology_landing_campaign_20260711/reports/FINAL_REPORT.md

教训2026-07-04

并发污染假阳:eval run 与实装 worker 并行导致 isolation 误报(教训入 ledger)

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/production_readiness_20260704/FINAL_REPORT.md

复发簇 · ledger_impersonation

2 次
教训2026-07-11

S4 台账冒名:worker 把「主 agent 亲验」写进我的台账

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/methodology_landing_campaign_20260711/reports/FINAL_REPORT.md

教训2026-07-04

S4 台账冒名:worker 把「主 agent 亲验」写进我的台账

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/production_readiness_20260704/FINAL_REPORT.md

复发簇 · pseudo_cross_domain_edges

2 次
教训2026-07-11

「5 条伪跨域边」叙事订正:先前 STATE/REBOOT 写的头条在磁盘产物中不成立

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/methodology_landing_campaign_20260711/reports/FINAL_REPORT.md

教训2026-07-11

campaign 叙事「5 条伪跨域边」头条在磁盘产物中不成立——这是本 campaign 自身要治的病(报表向上漂/账本向下漂)的一个现场实例

展开

根因:

对策:

adhoc_jobs/context_infra_base_tooling_buildout_20260615/methodology_landing_campaign_20260711/impl/rollout/FINAL_EVAL.md

复发簇 · regulator_missing_task_id

2 次
错误台账resolved2026-07-04

requirement_analysis_gate --regulator-tier codex:high 首跑 FLAG:顶层报 regulator_parse_error=no_json_object_found,真实原因在 runner 段 reason=missing_task_id(regulator_runner requires TODO_ID)。表层 parse_error 掩盖了缺任务锚这个真因,易误判为 codex…

展开

处置:带任务锚重跑:TODO_ID=<Txxxx> python3 tools/requirement_analysis_gate/check.py ...(intake/add 先拿到 todo id)。诊断顺序:先看 --json 输出 regulator.runner.reason,再怀疑回包格式

tool_failure · router_soft · requirement_analysis_gate.regulator

adhoc_jobs/idea_to_production_system_20260628/round4_cross_optimization_20260704/gate_run/source_manifest.tsv 同目录首跑输出(session b941545f)

错误台账resolved2026-07-04

phase 驱动器(Stop hook)自跑 gate_pass:check_report_completeness 时环境无 TODO_ID,regulator_runner missing_task_id 早退、顶层报 no_json_object_found(ERR-000006 在自动路径上的复发)

展开

处置:用 per-session active_task 锚(identity 解析第三顺位):python3 tools/tool_receipts/active_task.py set <Txxxx> --session-id <session>;驱动器 gate 子进程经 active_task 文件拿到 task_id。手动跑仍可 TODO_ID=<Txxxx> 前缀

tool_failure · router_soft · phase_driver.gate_pass.check_report_completeness

session b941545f Stop hook feedback 2026-07-04(P5 blocked)

复发簇 · tsprobe_w5x_20260706

2 次
错误台账resolved2026-07-06

TSPROBE W5-X controlled injection (NOT a real failure): synthetic tool_failure probe to verify the test system closes the five-segment loop (discover, register, schedule, fix, write back). Fix = mark this row resolved a…

展开

处置:TSPROBE W5-X fix segment 2026-07-06: probe defect repaired (probe evidence flipped to PASS); row resolved by the injecting experiment agent. No sanctioned resolve CLI exists yet; resolved via in-place ledger row update.

tool_failure · router_soft · w5x/tsprobe_true_discovery_probe

adhoc_jobs/systems_consolidation_20260705/runs/r028_w5x_true_discovery_probe/probe_evidence/defect/a1_det_verdict.json

错误台账resolved2026-07-06

resolution update for ERR-000067: W5M replay: sanctioned resolve CLI replaced the W5X manual JSONL flip; append-only correction verified

展开

处置:W5M replay: sanctioned resolve CLI replaced the W5X manual JSONL flip; append-only correction verified

tool_failure · router_soft · w5x/tsprobe_true_discovery_probe

adhoc_jobs/systems_consolidation_20260705/runs/r030_w5m_scheduling_bridge/REPORT.md

复发簇 · cron_daemon_duplicate_lock

2 次
错误台账unresolved2026-07-07

crontab 写入持续挂死(读正常):管道式与文件式 crontab - 均无限阻塞;ps 见 3 个 /usr/sbin/cron root 守护并存(Fri 常驻+今晨 8:00/8:30 各新增一个),疑 /var/at 锁争用。同 session 08:08 的 lock_watchdog 行安装成功,8:30 后全部写入挂死

展开

处置:未解:杀 root cron 守护需 sudo 超自主档位;crontab 内容已核完好;待用户手动装一行或重启 cron 服务后重试

environment · silent · wave3/merge 仲裁 cron 行安装

ps aux(三 cron 进程)+ ~/.cache/context-infra/crontab_backup_20260707_084317.txt(crontab 本体完好 121 行)

错误台账resolved2026-07-07

resolution update for ERR-000096: 根因修正:非 cron 守护重复,是 macOS TCC 权限拦 crontab 写入(静默挂死无提示)。用户在场批准权限后 crontab /tmp 文件式安装即通(INSTALL-OK,124 行=原121+3,lock_watchdog+仲裁行共 3 命中)。「3 个 /usr/sbin/cron」复判为 cron 执行整点任务的正常子进程(8:00/8:30 …

展开

处置:根因修正:非 cron 守护重复,是 macOS TCC 权限拦 crontab 写入(静默挂死无提示)。用户在场批准权限后 crontab /tmp 文件式安装即通(INSTALL-OK,124 行=原121+3,lock_watchdog+仲裁行共 3 命中)。「3 个 /usr/sbin/cron」复判为 cron 执行整点任务的正常子进程(8:00/8:30 与任务时刻吻合)。复用要点:crontab 写入挂死先怀疑 TCC,让用户在场批权限重跑,勿杀 root 进程

environment · silent · wave3/merge 仲裁 cron 行安装

crontab -l 124 行 + grep 3 命中

复发簇 · auto_regulator_runner:cannot_import_name_UnionType_from_types_Library_Developer_CommandLineTools_Libra:claude-kimi:medium:ImportError

2 次
错误台账unresolved2026-07-13

regulator_runner cannot_import_name_UnionType_from_types_Library_Developer_CommandLineTools_Libra: cannot import name 'UnionType' from 'types' (/Library/Developer/CommandLineTools/Library/Frameworks/Python3.framework/Ve…

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

tool_failure · router_soft · regulator_runner/DIAGDDP/cannot_import_name_UnionType_from_types_Library_Developer_CommandLineTools_Libra

tools/infra_core/regulator_runner.py:run_regulator

错误台账unresolved2026-07-13

regulator_runner cannot_import_name_UnionType_from_types_Library_Developer_CommandLineTools_Libra: cannot import name 'UnionType' from 'types' (/Library/Developer/CommandLineTools/Library/Frameworks/Python3.framework/Ve…

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

tool_failure · router_soft · regulator_runner/DIAGFIX39/cannot_import_name_UnionType_from_types_Library_Developer_CommandLineTools_Libra

tools/infra_core/regulator_runner.py:run_regulator

复发簇 · sandbox_sigterm_wrangler

2 次
错误台账workaround2026-07-22

Claude Code Bash 工具沙箱内跑 node tools/publish.mjs 三次被 SIGTERM(rc=143:两次死于 wrangler 上传步、一次死于内部 build 步 16 秒处);同命令 tmux 内三次全成功、真 cron 环境(career 每日链)一直正常。发送者未定位

展开

处置:wrangler 批量上传类命令一律在 tmux 或真 cron 跑,不在 CC Bash 沙箱前台跑;巡检链生产 runtime=cron 不受影响。若复发到 cron 面再深挖 sender

environment · silent · r005 publish 步(CC Bash 沙箱内)

runs/r005_dashboard_redesign_20260721/STATE.md 01:06 checkpoint + /tmp/r005-orig.log(ORIG_RC=143)

错误台账resolved2026-07-22

真 cron 环境复现 node 包装层 SIGTERM:PATROL_STATE 02:28:24 RUN_FAIL publish rc=143(BUILD_OK 后 34 秒被杀),推翻 ERR-000162『生产 cron 不受影响』定性;同窗口 claude 引擎在 cron 下 Not logged in rc=1(keychain 锁于无 UI 会话)耗 4 分钟后由 claude-kimi 正确接棒

展开

处置:①部署步弃 node 包装改 patrol.sh bash 原生(career deploy_verified 同款:npx 命令替换捕获+强制 Deployment complete 行+DEPLOY_RECORD bash 追记+hash 幂等跳过),publish.mjs 留给手动/tmux ②引擎链重排 claude-kimi→claude-zai→claude(env key 优先,订阅 CLI 降末位)③02:50 临时 cron 复测验证

environment · silent · r006 巡检 cron 实测 02:15

contexts/runtime/infra_status/PATROL_STATE.md 02:15-02:28 段 + logs/engine_20260722_021732_claude.log

单次条目(57)
教训2026-07-19

教训:teammate transcript 批量延迟落盘,mtime 停滞不是死亡信号——判失联应以交付消息为准或再等一个窗口

展开

根因:

对策:判 worker 失联以交付消息为准或再等一个窗口,不看 mtime

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r002_site_overhaul_20260719/reports/FINAL_REPORT.md

教训2026-07-20

空壳图防线:图存在 ≠ 图有内容

展开

根因:

对策:验收必须有密度探针

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r003_site_remediation_20260720/reports/FINAL_REPORT.md

教训ERR-0001622026-07-21

Claude Code Bash 沙箱内跑 wrangler 批量上传会被 SIGTERM(143)(遗留 R-1)

展开

根因:

对策:沙箱外(tmux/cron 谱系)跑 wrangler

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/runs/r005_dashboard_redesign_20260721/reports/FINAL_REPORT.md

教训2026-07-19

router 静默降级会破坏 provider-diversity 承诺

展开

根因:

对策:钉链+receipt 对账应成派发纪律

adhoc_jobs/context_infra_base_tooling_buildout_20260615/reflection_decisions_visibility_20260719/synthesis/FINAL_REPORT.md

错误台账unresolved2026-07-03

Run4 A4-F062 todo_chain_crons: tools/todo/review_domains_cron.sh lacked bounded parent timeout around todo.py review-domains router classifier; A4 run hung in codex:high domain review and was interrupted with rc=130, so…

展开

处置:未登记

tool_failure · silent · Run4/E_organic_dogfood/A4-F062-todo_chain_crons

adhoc_jobs/context_infra_base_tooling_buildout_20260615/Run4/rcl/A4_arming.jsonl:7; adhoc_jobs/context_infra_base_tooling_buildout_20260615/Run4/workstreams/A4_arming/evidence/todo/v5_review_domains.out

错误台账workaround2026-07-04

dao_shu_matrix_gate DM4 regulator(codex:high)内容侧 103 条 classification 全 consistent、counterexamples=[],但回包未套中 gate 期望的响应契约外壳:reason=custom_response_contract / runner_status=RAW_DELIVERED,fail-safe 整体判 FLAG(receipt TR-112…

展开

处置:内容实质 consistent 时复跑一次通常可得干净 PASS;根治需 regulator_runner 对 custom_response_contract 类回包做契约重试/宽容解析(工具债,未修)

tool_failure · router_soft · dao_shu_matrix_gate.regulator_wrapper

adhoc_jobs/idea_to_production_system_20260628/round4_cross_optimization_20260704/linkA/SUMMARY.md §4 gate 收口节

错误台账workaround2026-07-04

phase_skill_receipts 用 cursor.updated_ts 做时间截断,wait marker set/clear 更新 cursor 后同 phase 内更早的合法 skill_read receipt 被判 MISSING(静默假阴性,靠人工比对 receipt 台账才发现)

展开

处置:头部重读 skill 文件补新 receipt;根治=cutoff 取 phase advance 时刻,挂 BUG-R5-09 入 Run5 W0; receipt_id backfilled by tools/error_ledger/triage.py: TR-121457 score=90 reason=task_id+chain_step+time<=428s

tool_failure · silent · run5/P1/receipt-check

tools/phase_state/driver.py:180; receipts.jsonl TR-120914 (17:47:49, phase 标签全对) vs status MISSING @17:52

错误台账workaround2026-07-04

并发三闸核查发现磁盘 avail=0G(Data 卷 100%):.git 膨胀到 29G,其中 16.7G 为不可达松散对象(批量 add 失败/rc=124 中止提交的孤儿 blob,12.6k 个、均 ~1.4MB),另 143 个活 sharedindex×28M+434 session index=6.6G 常驻累积

展开

处置:应急=prune-packed + repack -d + prune --expire=1.hour.ago(14G 回收,1h 保护窗防在飞提交);根治挂 W0-C(BUG-R5-01 每 turn 400 文件提交+stderr 吞)与新 BUG-R5-11(index 无过期策略); receipt_id backfilled by tools/error_ledger/triage.py: TR-120943 score=80 reason=task_id+chain_step+time<=2910s

environment · silent · run5/P4/dispatch-gate

git count-objects 前后:loose 21102/17.68GiB → prune-packed+repack+prune(expire=1h) → 1670/2.59GiB;df 0G→15G;session-indexes 2.6G+sharedindex 148 个

错误台账resolved2026-07-04

W0-A 修复验证:router 硬 deadline+degenerate turns=0 归类 error;R5-04 由 stub 测试+一次真实 2400s 截断双证

展开

处置:landed via coordinator commit

tool_failure · router_soft · run5/W0-A

commit 待记; Run5/run_artifacts/W0-A/RESULT_COORDINATOR.md; 145 tests

错误台账workaround2026-07-04

Native 'claude -p' probe exited 1 with 'Not logged in · Please run /login'; project UserPromptSubmit hook could not be exercised through native Anthropic subscription.

展开

处置:Switched verification to claude-kimi wrapper (same CC binary, Kimi Coding Plan subscription); UserPromptSubmit hook fired and capture/kill-switch both verified.; receipt_id backfilled by tools/error_ledger/triage.py: TR-124944 score=80 reason=task_id+chain_step+time<=747s

provider_auth · router_soft · Run5/W0-D1/real_cc_probe/native_claude

adhoc_jobs/context_infra_base_tooling_buildout_20260615/Run5/run_artifacts/W0-D1/cc_probe_stdout.txt; cc_probe_exit.txt

错误台账workaround2026-07-04

cli_agent router hangs with zero stdout/stderr for claude:low, claude-zai:low (subprocess_timeout, TTY-park class), and ALSO times out for opencode:low and codex:low canaries while a concurrent codex:high worker runs fi…

展开

处置:S1 arms bypass router: direct codex exec with isolated workdir; router issue deferred to owner

provider_silent_failure · router_soft ·

scratchpad canary3.json/canary_oc.json empty rc=124; tasks/b0gy5c4ip.output shows classifier_verdict=subprocess_timeout proof_level=FAILED

错误台账resolved2026-07-04

W0-A修复验证: codex-oc:high声明subprocess_timeout=1800实跑4453s未被截断(turns=0消耗约6M token仍判[OK]被调用方is_error吞掉); codex-exec/claude两条subprocess路径超时被误归possible_permission_flag_hang; dispatch超deadline历史上不返回需手动kill

展开

处置:8e5418c60把subprocess_timeout透传进_invoke_opencode+_RouterDeadlineExceeded硬截断+turns=0判degenerate_success; 本轮加codex-exec/claude两路_looks_like_subprocess_timeout归类subprocess_timeout+error路径跳过约70s的canary扫描避免重蹈deadline失效; 魔法数集中SUBPROCESS_TIMEOUT_GRACE_S

tool_failure · router_soft · run5/W0-A/router

run_artifacts/W0-A/RESULT.md; router.py:1845/2049/2236/2256/2363; dispatch.py:441; 负例(全路径3.01s不截断/claude误归类possible_permission_flag_hang)→正例(1.10s截断[FAIL]/subprocess_timeout)

错误台账workaround2026-07-05

python3 -m tools.phase_state.cli arm wrote phase cursor for codex-019f32eb-7c42-7e83-8a38-0cb0eb84455a but produced no final output for over 120s; KeyboardInterrupt stack showed fcntl.flock(handle.fileno(), fcntl.LOCK_E…

展开

处置:Workaround: interrupt the stuck receipt write after confirming phase_cursor exists with phase_state status; continue with cursor-backed phase state. Root fix remains H-1 receipt ledger governance/sharding.

tool_failure · silent · SES-3/P1/phase_state_arm

this session tool output: phase_state arm KeyboardInterrupt stack; contexts/tool_receipts/receipts.jsonl size 145M

错误台账workaround2026-07-05

phase_state arm created the phase cursor but hung while writing phase_cursor_armed receipt at tools/tool_receipts/receipt.py fcntl.flock LOCK_EX

展开

处置:Do not forge missing Codex skill receipts; use run-local artifacts and phase_markers.md for mechanical evidence in this session

tool_failure · silent · r004/P1/phase_state_arm

错误台账workaround2026-07-05

todo.py start T1230 completed state/event write but hung on tools/tool_receipts/receipt.py fcntl.flock(...) until interrupted; subsequent todo writes used isolated TOOL_RECEIPTS_LEDGER.

展开

处置:Use isolated TOOL_RECEIPTS_LEDGER for todo commands in this session; T1230 state/event had already been emitted; root fix remains the receipt ledger O(1)/locking slice.

tool_failure · silent · T1230/todo_start/receipt_record

tools/todo/todo.py:2846-2861; tools/tool_receipts/receipt.py:196; contexts/todo/logs/todo_events.jsonl:254; adhoc_jobs/systems_consolidation_20260705/runs/r005_20260705_ses1_hook_root_fix/RUN.md

错误台账workaround2026-07-05

Running python3 tools/tool_receipts/receipt.py init-seq failed with ModuleNotFoundError: No module named 'tools' because direct script execution did not put the repo root package on sys.path.

展开

处置:Use python3 -m tools.tool_receipts.receipt init-seq from repo root, or add a future direct-script sys.path shim if file execution is required.; receipt_id backfilled by tools/error_ledger/triage.py: TR-153793890 score=70 reason=task_id+time<=2s

tool_failure · silent · r005/P2/init_seq_cli

tools/tool_receipts/receipt.py direct CLI attempt in r005 session

错误台账resolved2026-07-05

Initial receipt concurrency benchmark used multiprocessing from python - <<PY; Python 3.14/macOS spawn could not import <stdin> and emitted FileNotFoundError for <stdin>.

展开

处置:Use subprocess.Popen with python -c child code for concurrent hook-load benchmarks; do not use multiprocessing Pool from stdin scripts on this runtime.; receipt_id backfilled by tools/error_ledger/triage.py: TR-153793890 score=80 reason=task_id+domain+time<=2s

tool_failure · silent · r005/P5/receipt_benchmark

r005 terminal output; rerun used subprocess python -c workers

错误台账resolved2026-07-05

dispatch_edge first dogfood inherited session_dispatch router fallback_chain: codex:high failed subprocess_error, then Kimi timed out, OpenCode create_session_failed, and Zai kept waiting; this exceeded the queue timeou…

展开

处置:Patch dispatch_edge to pass fallback_chain=[] into session_dispatch.dispatch_helper so consume uses the queue executor.channel only; rerun dogfood after resetting/choosing a pending task.

provider_routing · router_soft · T1264/dispatch_edge/dogfood_attempt_1

/tmp/dispatch_edge_t1264.YV6Wwv/W2-T1234-SCOREBOARD-TRIGGER-DOGFOOD/2026-07-05T192739_0200/session_dispatch/router.log

错误台账unresolved2026-07-05

Nested codex exec cannot initialize in this managed Codex sandbox: direct codex --ask-for-approval never exec exits rc=1 with failed to initialize in-process app-server client: Operation not permitted. This blocks the r…

展开

处置:No in-session permission escalation is available (approval policy never). Keep dispatch_edge strict codex behavior, mark real codex dogfood blocked, and rerun from an unrestricted/current app-server-capable session.

environment · http_hard · T1264/dispatch_edge/codex_probe

/tmp/dispatch_edge_codex_probe.txt (absent because codex init failed); terminal stderr in T1264 session

错误台账resolved2026-07-05

todo defer-verify failed because --deferred-class implementation is not a valid enum

展开

处置:reran todo defer-verify T1205 with --deferred-class doc_only; command succeeded and status became awaiting_verify

tool_failure · router_soft · r007/todo/defer_verify

adhoc_jobs/systems_consolidation_20260705/runs/r007_w2a_bugtrace_ddp/FINAL_REPORT.md#todo-status

错误台账deferred2026-07-05

W3-B Q2 OpenCode verification found deployed plugin drift from repo source and 4096 attach canary did not pass: first run could not open ~/.local/share/opencode/log/opencode.log under sandbox; retry with run-local XDG_D…

展开

处置:No live user-level plugin write was performed from sandbox. Report contains rollback-capable backup/deploy/verify/canary/rollback package; acceptance remains hash match plus 4096 canary returning W3B_OPENCODE_CANARY_OK.

tool_failure · router_soft · T1173/W3B-Q2/opencode_4096_canary

adhoc_jobs/systems_consolidation_20260705/runs/r011_w3b_dc_opencode/OPENCODE_Q2_REPORT.md

错误台账unresolved2026-07-06

W5-B validation: controlled auto error should chain_emit into todo intake with agent_finding source_kind.

展开

处置:未登记

tool_failure · router_soft · w5b/auto_emit_chain_probe

adhoc_jobs/systems_consolidation_20260705/runs/r017_w5b_ledger_liveness/REPORT.md#auto-chain-probe

错误台账resolved2026-07-06

W5G1 kimi worker hit billing-cycle quota (403) after 65 turns/553s mid-implementation; lane died on final viable tier with no fallback chain

展开

处置:requeue lane on claude-zai:high with continuation note; kimi removed from tonight pool; live justification for T1273 disciplined fallback

provider_quota · http_hard ·

/tmp/dispatch_edge_runs/W5G1-TSM-SLICES/2026-07-06T011944_0200/session_dispatch/W5G1-TSM-SLICES_result.json api_error_status=403

错误台账resolved2026-07-06

dark_context_s3_loop gate passed by direct CLI but failed phase_runtime gate_pass projection because unified_gate appends --skip-regulator and the new CLI did not accept it.

展开

处置:Added a no-op top-level --skip-regulator compatibility flag to tools/dark_context/s3_loop_gate.py and reran phase_runtime gate_pass successfully.

tool_failure · http_hard · r024_w5h_dc_s3/phase_runtime_projection

adhoc_jobs/systems_consolidation_20260705/runs/r024_w5h_dc_s3/s3_real_task_skill_routing/s3_loop_gate_report.md

错误台账workaround2026-07-06

T1172 pressure evidence attempted global ps process count, but current sandbox denied ps with operation not permitted; switched to launched-PID live count evidence.

展开

处置:Use subprocess Popen handles for controlled bridge PID live-count pressure evidence; keep global ps count unavailable noted in REPORT.

environment · http_hard · W5K-RUN5-RESIDUAL-HIGH/T1172/pressure-ps-count

错误台账deferred2026-07-06

T1153 codex-oc:high 30s canary resolved to opencode model_id_configured=gpt-5.5, but HTTP session creation to localhost:4096 failed with Operation not permitted in current sandbox.

展开

处置:Record failed coverage attempt; true hit remains unavailable until localhost opencode access is permitted or the canary is run from an unsandboxed host.

environment · http_hard · W5K-RUN5-RESIDUAL-HIGH/T1153/codex-oc-high-canary

错误台账deferred2026-07-06

T1146 controlled SINCE_DAYS=14 codex_rescue_cron ran for the 14-day window but exited rc=128 with 'unable to create temporary file: Operation not permitted' while adding to the git object database.

展开

处置:Do not widen to BUG-016 phase C or force writes in this sandbox; keep lock regression and doctor monitor landed, report backfill blocked by environment.

environment · http_hard · W5K-RUN5-RESIDUAL-HIGH/T1146/SINCE_DAYS_14_codex_rescue_cron

错误台账resolved2026-07-06

W5K initially ran multiple todo.py write operations in parallel; later show proved T1172/T1157 status writes were lost while T1177 persisted.

展开

处置:Rerun todo status mutations sequentially and avoid parallel writes to contexts/todo/TODO.md.

tool_failure · silent · W5K-RUN5-RESIDUAL-HIGH/todo-status-update

错误台账deferred2026-07-06

W5K explicit git add of whitelist files failed with fatal: Unable to create .git/index.lock: Operation not permitted. The workspace allows file edits but this Codex sandbox cannot write the repository git index.

展开

处置:All code/report/todo/RCL changes are on disk but not committed to HEAD; commit must be performed from a session with .git write permission.

environment · http_hard · W5K-RUN5-RESIDUAL-HIGH/git-stage-commit

错误台账deferred2026-07-06

T1284 attempted to write ~/.claude/settings.json after creating backup and normalized after-copy, but Codex managed sandbox denied write with PermissionError Operation not permitted.

展开

处置:Live user-level settings install must be run from a non-sandbox shell: cp r029 backups/settings.json.after ~/.claude/settings.json; rollback cp settings.json.before back.

environment · http_hard · r029_w5l1_hook_capture/T1284_user_settings_write

terminal: PermissionError Operation not permitted writing ~/.claude/settings.json; normalized file at adhoc_jobs/systems_consolidation_20260705/runs/r029_w5l1_hook_capture/backups/settings.json.after

错误台账deferred2026-07-06

python3 tools/git/pending_merges.py compress --path ~/.cache/context-infra/pending_merges.md failed writing ~/.cache/context-infra/pending_merges.md.tmp.* with PermissionError: [Errno 1] Operation not permitted

展开

处置:Backup was created in run directory; live compression must be replayed from a main environment with HOME cache write permission.

environment · http_hard · W5P/T1462/pending_merges_compress

adhoc_jobs/systems_consolidation_20260705/runs/r033_w5p_git_nightwindow/pending_merges.md.before_t1462_20260706T051329Z.bak

错误台账deferred2026-07-06

git merge-tree --write-tree --messages <main> <refs/sessions/*> failed with error: unable to create temporary file: Operation not permitted; fatal: failure to merge

展开

处置:Router unit tests passed, but real merge-tree dry-run must be replayed from an environment with .git object write permission.

environment · http_hard · W5P/T1464/merge_tree_dry_run

adhoc_jobs/systems_consolidation_20260705/runs/r033_w5p_git_nightwindow/t1464_three_ref_dry_run.jsonl

错误台账deferred2026-07-06

T1146 apply precheck found doctor ok=true but ps unavailable, orphan git gateway lease suspected, breaker liveness unverifiable, concurrent refs/objects changed before apply, and merge-tree/hash-object writes are denied…

展开

处置:Did not run SINCE_DAYS=14 apply in this sandbox; replay from main environment with .git write permission and clean writer/lease precheck.

environment · http_hard · W5P/T1146/codex_rescue_apply_precheck

adhoc_jobs/systems_consolidation_20260705/runs/r027_w5k_run5_residual/CODEX_RESCUE_SINCE14_SUMMARY.md

错误台账deferred2026-07-06

W5G2-RESUME2 dispatched on claude-zai:high but died turn-1 with KIMI 403 quota error - channel misroute or wrapper env crossover

展开

处置:products harvested from worktree by coordinator; misroute investigation folded into T1273/T1475

provider_routing · router_soft ·

/tmp/session_dispatch_runs/W5G2-TSM-SEMANTIC-RESUME2*/W5G2*result.json shows kimi.com quota message under zai dispatch

错误台账unresolved2026-07-06

commit_wrapper failed while staging r031 W5-G2 report: fatal: Unable to create '.git/index.lock': Operation not permitted; git add skipped path and reported nothing to commit

展开

处置:Unresolved in this Codex sandbox because .git writes are denied; rerun the same commit_wrapper command from a main environment with .git write permission, staging only r031_w5g2_tsm_semantic/

environment · http_hard · r031_w5g2_tsm_semantic.commit_report

adhoc_jobs/systems_consolidation_20260705/runs/r031_w5g2_tsm_semantic/COMMIT_BLOCKER.md:9

错误台账workaround2026-07-07

reference_validator.py rename --full-repo 移动文件名为 - 的文件时,CASCADE 把正则字符类 [\w./-] 中的 - 误判为文件引用并替换成归档路径,静默改坏 3 个现役 gate 工具正则(f2_number_traceability/filter_compress_gate/reanchor_gate)+1 build log+1 daily record;无报错,靠 worker…

展开

处置:git checkout -- 回滚 4 tracked 文件+手工回滚 untracked;历史审计文档改回原名加归档旁注;FILE_SYSTEM_OPERATIONS_GUIDE §3.2 加检查清单条目;防呆票 AF-REFVAL-CASCADE(rename 对单字符/纯符号文件名默认拒绝或强制人工确认)

tool_failure · silent · r003_landing_push/W1-F/D7 debris 归档

adhoc_jobs/redundant_archive/root_debris_20260707/ + DISPATCH_LOG.md(R-F 回滚核验干净)

错误台账resolved2026-07-07

route_on_prompt stdin 探针不带 session_id 时落 unknown-session dedup 桶;桶饱和后 hook 静默零输出,探针者会把 dedup 假象误判为路由失败(R-A 首测即踩中)

展开

处置:施测探针必须带全新 --session-id 且隔离 SKILL_ROUTER_DEDUP_DIR=$(mktemp -d);该约定已写进后续所有 regulator prompt 模板

tool_failure · silent · r003_landing_push/R-A regulator 施测

tools/skill_router_hook/route_on_prompt.py(dedup 逻辑)+ DISPATCH_LOG.md R-A 横切#5

错误台账unresolved2026-07-07

regulator_runner subprocess_timeout: subprocess_timeout

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

provider_silent_failure · silent · regulator_runner/T3054/subprocess_timeout

tools/infra_core/regulator_runner.py:run_regulator

错误台账workaround2026-07-08

After the user requested Test System upgrade and residual handling, the agent continued into Base Tooling Buildout FR-12 auto final report work instead of staying anchored to adhoc_jobs/test_system_landing_methodology_2…

展开

处置:Stop Base Tooling path for this goal; re-anchor next work to adhoc_jobs/test_system_landing_methodology_20260704 and only use Base Tooling context when explicitly requested or as comparison evidence.

orchestration_failure · silent · 2026-07-08/r007-to-r008/scope-selection

commits:50184e12d,80cd6ccab; user correction: Test System not Base Tooling

错误台账workaround2026-07-08

clause_status.py --now 2026-07-08T11:45:00+02:00 did not complete under an outer 300s timeout while rendering the Test System clause status report; KeyboardInterrupt stack showed serial pytest predicate evaluation in cl…

展开

处置:Work around for r7 by using the existing corrected CLAUSE_STATUS.generated.md and running targeted tests directly. Follow-up should cache duplicate predicate results or batch pytest node ids so report rendering has an overall bounded runtime.

tool_failure · silent · r7/clause_status_render

command: timeout 300 python3 tools/test_health/clause_status.py --now 2026-07-08T11:45:00+02:00 -> rc=124; stack: tools/test_health/clause_status.py:evaluate_clause -> clause_predicates._check_test_passes

错误台账workaround2026-07-08

r7 Test System provider review router attempts for codex:xhigh, claude:high, and claude-ollama-glm:high each timed out with rc=124 and no stdout/log evidence; direct codex exec succeeded separately for the primary GPT-5…

展开

处置:Use direct codex exec as primary bounded verdict; record Opus/GLM support as unavailable caveats and keep multi-provider solid decision incomplete.

provider_silent_failure · silent · r7_testsys_replacement/provider_review

adhoc_jobs/test_system_landing_methodology_20260704/runs/r7_testsys_replacement_20260708/logs/PROVIDER_ROUTE_FAILURES.md

错误台账deferred2026-07-08

Targeted local verification for tools/test_health/tests/test_clause_status_s2.py timed out at 180s after emitting 21 passing dots and no failure diagnostics; combined clause_status_s2+diff_table pytest also timed out at…

展开

处置:Keep diff_table verification as passed; record clause_status_s2 timeout as a Test System health residual pending predicate batching/caching or narrower node selection.

tool_failure · silent · r7_testsys_replacement/local_verification

adhoc_jobs/test_system_landing_methodology_20260704/runs/r7_testsys_replacement_20260708/logs/TARGETED_PYTEST_TIMEOUT.md

错误台账resolved2026-07-09

context-infra disk refilled rapidly because contexts/tool_receipts/receipts.jsonl (~238-254MB append-heavy receipt log) was still tracked and Codex App repeatedly ran git add -A/write-tree, producing ~56-57GiB loose Git…

展开

处置:Ran git prune --expire=now and git gc --prune=now; .git dropped to ~4.7G, loose objects to 63 / 860KiB, system free space recovered to ~73GiB; added contexts/tool_receipts/receipts.jsonl to .gitignore and removed it from Git index with git rm --cached while preserving the local file.

environment · silent · 20260709_git_object_cleanup

contexts/survey_sessions/disk_space_full_diagnosis_20260707_manual.md:276

错误台账unresolved2026-07-10

regulator_runner expectation_bias_detected: regulator question carries expectation-direction language: question matched expected_verdict: '预期内容、要给测试 session 一套工具+skill 操作指南 | `workflow_buildout_real_session_eval`(道·评估方法…

展开

处置:Inspect regulator attempt receipts and rerun from the main environment if needed.

tool_failure · router_soft · regulator_runner/MIG-REB-P2/expectation_bias_detected

tools/infra_core/regulator_runner.py:run_regulator

错误台账resolved2026-07-18

background Agent(afac-v3) 01:18 启动 run.py fin域重跑后 01:22 转 idle; 重跑 01:25 exit 0 完成但 agent 33min 无步骤4/5 产出(v3_report.md 不存在, v3_progress.md 停在 01:18), 其 sleep 轮询进程消失; 无任何错误码

展开

处置:Orchestrator SendMessage 催醒(01:59), agent 4min 内复活并完成收口(02:03 报告落盘)。预防: 派发 prompt 显式要求长跑收口用同 turn blocking 轮询(until 循环)而非启动后即结束 turn; Orchestrator 心跳按产物 mtime 判 stall 并催醒, 两次无响应才重派

orchestration_failure · silent · afac_v3_finalization

~/code/afac_task4/logs/v3_progress.md:01:18条目与02:03条目间33min空窗

错误台账resolved2026-07-18

前任 v3 实现 session 工作至 07-18 00:46 被终止(额度窗口耗尽), 无 handoff 无进度日志遗留; 留下未提交 WIP(answer.py+49/table_facts.py+227) 与基于未验证错误事实的 out/ 重跑结果, 且 stash@{0} 另有更早半成品混淆现场

展开

处置:新 Orchestrator 冷启动后 git 取证(diff/stash show -p md5 对比/文件 mtime 时间线)判 WIP 有价值, 派 Fable 续跑 agent 验证优先增量续跑(锚点核验抓出 4 类解析 bug, 证明 00:46 结果不可用)。预防: 实现 agent 硬性要求每完成一步 append 进度日志(断点恢复配方); Orchestrator 维护 RUN_STATE 重派配方 + 5h 额度保险(cron+watchdog)

provider_quota · silent · afac_v3_initial_implementation

~/code/afac_task4 git diff/stash md5 对比 + out_v2(00:40) vs out(00:46) mtime 三角定位; DECISIONS.md §2 07-18 条目

错误台账resolved2026-07-18

git_guard.sh 危险操作分类器 import 链在 tools/infra_core/clean_context.py:28 (from types import UnionType, Python 3.10+ 语法) 撞系统 /usr/bin/python3=3.9.6 墙抛 ImportError,触发 GUARD_FAILURE_ALLOW 静默 fail-open;07-13~07-18 累计 79 次,事发窗口 2…

展开

处置:config/bin_paths.sh 新增 GIT_TOOLING_PYTHON3 解析器(探测 >=3.10 + fcntl 双重条件,优先 /opt/homebrew/bin/python3);git_guard.sh PYTHON_BIN 改用同构探测函数 + classify_operation import 加 try/except 按文件路径兜底加载(跳过 tools.orchestrator 包 __init__ 的 phase_runtime 依赖链,classifier 本体 stdlib-only 可在 3.9 独立跑)

tool_failure · silent · git_guard/danger_op_audit/classify_operation

contexts/runtime/orchestrator/danger_op_audit.jsonl (GUARD_FAILURE_ALLOW tier=UNKNOWN, ImportError traceback); tools/git/hooks/git_guard.sh:20 (PYTHON_BIN 硬编码 /usr/bin/python3); contexts/survey_sessions/git_tracking_anomaly_forensics_20260718_manual.md §2

错误台账resolved2026-07-18

共享checkout(多session并发写同一工作树)内,两个不同session的独立sub-agent(cursor-integrator于01:23、w-todo-export于01:46-01:48)在25分钟内各自独立复发'用git stash当干净测试基线'反模式;w-todo-export一轮557文件+56896/-52560行全仓未提交工作被扫进stash@{0}、工作树回退,波及4+并发session在途工作(pr…

展开

处置:现场恢复=以'工作树==HEAD才恢复'判据从stash@{0}批量restore --source(sanctioned模式,非pop),中文文件名quotepath坑+index.lock竞争两轮补跑,470文件零残留;结构修复=danger_op_audit.py新增DangerAuditContext.in_shared_checkout字段+_stash_kind()三态分类(read/scoped_push/sweep),共享checkout内sweep归档C(HUMAN_GATE)、scoped_push归档B(safety tag);git_guard.sh危险候选集与fail-closed兜底清单均补stash家族;rules/git_safety.md §1补充sanctioned替代方案(git worktree add临时只读基线/带--pathspec窄stash/git restore --source恢复不用pop);dispatch层面的'临时只读基线'工具化配方尚未落地(登记OPEN_PROBLEMS.md GIT-3)

orchestration_failure · silent · worker_dispatch/shared_checkout/clean_baseline_testing

session 941444e4/9ddc6821 sub-agent raw transcript tool_use 记录(23:23:16Z/23:46:08Z/23:47:41Z/23:47:59Z 原始stash命令); git reflog(reset: moving to HEAD 01:23:29/01:46:29/01:48:09); stash@{0} 对象时间戳与--stat(557 files); contexts/survey_sessions/git_tracking_anomaly_forensics_20260718_manual.md §1 事件B/C + §4 处置建议3

错误台账resolved2026-07-18

两处独立问题在同一测试环境修复轮次(2026-07-18)被一并处理: (1) tools/cli_agent/router.py:205 与 aau_v2/runners.py:85 的 LLM_CODEX_POLICY_CONTEXT_LIMIT 兜底默认值从272_000/240_000拆分更新为300_000(context_limit)+240_000(auto_compact_token_limit, 原200_000)—…

展开

处置:router.py/runners.py 的策略默认值已更新为300_000/240_000并与272_000 exec硬顶解耦(两者不再是同一数字, 语义上本就不该相同); test 套件加 importorskip 四处使3.9环境下优雅跳过而非整体崩溃; /opt/homebrew/bin/python3 -m pytest tools/cli_agent/tests/test_codex_effort_profiles.py 实测41 passed

tool_failure · silent · cli_agent/router/_codex_gpt56_policy

tools/cli_agent/router.py:205 (git diff); adhoc_jobs/atomic_agent_unit_20260509/second_generation_implementation_20260511/aau_v2/runners.py:85 (git diff); tools/cli_agent/tests/test_codex_effort_profiles.py (git diff, importorskip x4 处 + test_opencode_policy_derives_same_200k_trigger 更名为 test_opencode_policy_trigger_stays_inside_channel_windows 并拆分断言); tools/llm_runtime/providers.conf:129 (272000 exec catalog硬顶仍有效, live 2026-07-18)

错误台账resolved2026-07-19

Stop hook stop_phase_check.py blocked session c1f1275b with foreign phase plan: 'current phase information_search (P1-FREEZE-AND-LOCATE) is not complete... plan: adhoc_jobs/context_infra_migration_20260709/runs/audit_20…

展开

处置:Standard response (Orchestrator-ratified): 1) run phase_state.cli status --session <id> to identify the bound plan and its true owner; 2) if foreign: read required skills honestly (cheap), but DO NOT fabricate the plan's exit artifacts (writing completion markers for un-performed work is claim falsification); 3) escalate via lane's ORCH_REQUEST channel with plan path + purpose mismatch evidence; 4) owner side resolves with decline-unbind (verified: status shows 'declined, phase driver is silent'). Work continues throughout - Stop hook blocks only session termination, not execution.

orchestration_failure · router_soft · lane/P1-P6/stop_hook

~/code/afac_task4/logs/ORCH_REQUEST_phase_misarm.md

错误台账unresolved2026-07-19

stop hook synthetic staging repeatedly fails to git add .claude/settings.json; partial capture each turn, only ~/.cache text log records it

展开

处置:de-ignore or special-case the path in synthetic staging filter; see audit memo trace_base_tooling_state_audit_20260719_manual.md; todo T3923

tool_failure · silent · stop_hook/synthetic_staging/git_add

~/.cache/context-infra/stop_hook.log:2026-07-19T11:33:41 WARN synthetic git add failed for path=.claude/settings.json

错误台账unresolved2026-07-19

reflector cron (Sun 14:00, periodic_jobs/ai_heartbeat/src/v0/reflector.py) has refused OBSERVATIONS.md -> memory promotion for 7+ days straight (stalled since 2026-07-12, last successful promotion date:2026-07-11 ran-at…

展开

处置:Not yet fixed. Needs: (1) root-cause the isolated OpenCode recovery session exit 127 (likely PATH/interpreter issue in the isolated recovery subprocess, same family as other exit-127 interpreter-wall incidents this repo has hit before); (2) manually quarantine or clean-replace the 2026-07-15 and 2026-07-18 high-suspicion observation segments so pre-flight stops blocking; (3) once unblocked, verify reflector actually promotes the 7-day backlog (2026-07-12 through 2026-07-18 Date segments) rather than silently skipping them.

tool_failure · silent · reflector/preflight/isolated_opencode_recovery

/tmp/reflector.log (Jul 19 14:00 run, lines 1-15); contexts/observations/OBSERVATIONS.md (last reflected:1 marker at date:2026-07-11 ran-at:2026-07-12T13:22:46Z; grep -c '^Date: 2026-07-1[2-8]' = 7 unconsumed Date segments); adhoc_jobs/methodology_system_buildout_20260719/collection/c7_feedback_channels.md (row 1 of channel table + breakage item 1)

错误台账resolved2026-07-20

Dispatched general-purpose verifier sub-agent (a0ba691c) for v6 double-source check; agent reported 'scripts at scratchpad/verify_indep.py, output scratchpad/verify_out.txt' but those paths under project root did not ex…

展开

处置:Verified files DO exist (agent genuinely ran independent recompute — verify_indep.py 188 lines, does not import iter2_offline_sweep, rebuilds index independently; numbers match main sweep). Copied both to project-persistent scratchpad/verify_indep.py + verify_out.txt as durable anchors. Preventive: verifier-worker prompts MUST specify an absolute persistent output path under the project tree, not rely on sub-agent default cwd. Confirms E16 (methodology §5): session-ephemeral dirs must not serve as persistent evidence anchors in handoff.

orchestration_failure · silent · lane/ITER2/verifier_dispatch

logs/W5_ITER2_LAUNCH_REVIEW.md

错误台账resolved2026-07-22

cron 日志逐字:env: node: No such file or directory / /bin/sh: node: command not found / [publish] build 失败(致命):Command failed: node build.mjs——publish.mjs 内 execSync 裸调 node,cron 极简 PATH 无 nvm bin,自 07-20 起每日 08:15 连败

展开

处置:publish.mjs 子进程一律绝对路径:NODE=process.execPath、NPX=dirname(NODE)/npx;同类脚本(patrol.sh)头部显式 export PATH 含 nvm bin(抄 career run_digest.sh 同款)

environment · silent · infra-status publish cron 08:15

~/.cache/context-infra/site_publish.log + site_candidates/kimi/tools/publish.mjs:100

错误台账resolved2026-07-22

stdin 为 socket 时 wrangler 4.113.0 只打印版本横幅即退出 rc=0,连上传都未发生(deployment list 无新记录);同命令 tmux 真 TTY / cron /dev/null stdin 下正常。另 BSD script 伪 TTY 方案在 socket stdin 下直接失败:script: tcgetattr/ioctl: Operation not supported on soc…

展开

处置:废弃 script 伪 TTY;execSync 给 wrangler stdio:['ignore','pipe','inherit'] 钉 stdin + 输出强制含 Deployment complete 行才判 DEPLOYED(缺=FAKE_SUCCESS exit 3)+ 部署后带 ?v= 缓存穿透参数对生产域做 index 字节轮询(career deploy_verified 三件套同款)

provider_silent_failure · silent · r005 wrangler pages deploy(infra-status)

/tmp/r005-wrangler-direct.log + contexts/runtime/infra_status/logs/publish_20260722.log

错误台账resolved2026-07-22

前端 worker 交付的 build 断言③『密码门全页注入』实现为 if (written.length < 250) throw ——只数页数不检查门节点;负对照实测:把 shell.mjs 的 gate-card 改名后 build 仍 PASS(门整体消失也放行)

展开

处置:断言改为逐页核三个门标记(gate-card/sessionStorage/SHA-256)readFileSync 实读 dist 产物;负对照重跑 259 页全 FAIL 证其真能抓。通用纪律:验收断言必须能被负对照证伪,『数量达标』不等于『东西还在』(与 r004 空壳图密度探针同族)

tool_failure · silent · r006 前端交付验收

site_candidates/kimi/build.mjs:188-190(修前)+ /tmp 负对照两次构建日志

错误台账resolved2026-07-22

patrol.sh line 90: PUB_LOG): unbound variable —— die 消息里中文全角括号紧贴 $PUB_LOG,bash 把「)」当变量名一部分;03:55 整轮 cron 白跑(BUILD_OK 后即死)

展开

处置:错误消息一律 ASCII + ${var} 花括号形式;career run_digest.sh 2026-07-19 踩过同款(BUILD_STATE 有案),本次固化为纪律写入 lessons

environment · http_hard · r006 patrol.sh bash 原生部署步

~/.cache/context-infra/infra_status_patrol_cron.log 03:55 段