[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6PVD72FAGQDRD1EWHTEDWVK":3,"skill-same:01M6PVD72FAGQDRD1EWHTEDWVK":62},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":34,"linkOnly":35,"markdownContent":36,"markdownBody":37,"frontMatter":38,"refName":42,"commitSha":43,"skillDirectory":44,"forksCount":41,"repositoryHtmlUrl":45,"repositoryHomepageUrl":6,"ownerHtmlUrl":46,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":47,"install":48,"resources":61},"01M6PVD72FAGQDRD1EWHTEDWVK","self-retrospective",null,"Turn what went wrong into a systemic fix. Use when the supervisor asks for a retro, after a failed task (mandatory), after a task with backtracks, or when wrapping up a multi-task session. Not the automated audit that runs after a session closes.","muratovv\u002Fai-hats","muratovv","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F7502801?v=4","01M6JGCBN4KEJX012GPJVSH04T","Fedor Muratov","01M6JGCBSNMN7CW3RGTPPE0MN2",3,"MIT","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fcore\u002Fskills\u002Fself-retrospective\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32,33],"retrospective","self","self retrospective","audit","retro","supervisor","fix","session","use","up","task","multi","with","for","Wrong","2026-11-10 10:12:30",false,"---\nname: self-retrospective\ndescription: >-\n  Turn what went wrong into a systemic fix. Use when the supervisor asks\n  for a retro, after a failed task (mandatory), after a task with\n  backtracks, or when wrapping up a multi-task session. Not the automated\n  audit that runs after a session closes.\nlicense: MIT\nai_hats:\n  schema_version: 1\n  type: skill\n---\n\n# Self-Retrospective\n\nAnalyze completed or failed work to identify systemic improvements.\n\n## When to Use\n\n- When the supervisor explicitly asks to \"write a retro\" \u002F \"do a\n  retrospective\" — follow THIS protocol (4.5 filter, step-7 backlog gates);\n  do not free-form a retro file by pattern-matching an existing one.\n- At the end of a multi-task work session (wrap-up across several closed\n  tasks) — not only single-task completion.\n- After task completion (especially with failures or backtracks)\n- After a failed task (mandatory)\n- When the backlog manager transitions to `review` or `failed` state\n\n**Not** the automated session-review loop. `reflect-session`\n(`hats-reflect-session\u002Fv1`) is a pipeline the harness runs to triage\nhypotheses\u002Fproposals; `self-retrospective` is the personal agent reflection\nyou run by hand (free-form markdown, archived under `sessions\u002Fretros\u002F`). And\n**not** a factual outcome record — that is **task-summary**; this skill is the\n*improvement* analysis (why the work went as it did, what to change next time).\n\n## Procedure\n\n1. **Facts:** Chronologically list all errors, backtracks, wasted iterations,\n   or unexpected findings from the session.\n\n2. **Classify:** Assign each issue to a category:\n   - **Knowledge** — missing context or incorrect assumption about the system\n   - **Environment** — tooling, permissions, infrastructure issues\n   - **Process** — skipped steps, wrong order, missing verification\n   - **Communication** — misunderstood requirements, ambiguous instructions\n   - **Assumption** — untested belief that turned out wrong\n\n3. **Root Cause (5 Whys):** For each significant issue, ask \"Why?\" iteratively\n   until you reach an actionable root cause. Stop at the level where a fix\n   is practical.\n\n4. **Improvements:** Propose specific, concrete changes:\n   - Rule or skill updates (with exact content)\n   - Workflow changes (with steps)\n   - New checks or verification steps\n     Focus on **systemic** fixes over one-off patches.\n\n4.5 **Hypothesis candidates:** Surface improvements that describe a\n**behavioural pattern** (the agent systematically does X) — those go\ninto the hypothesis backlog, not the task tracker. The supervisor\nchooses what to record.\n\n    Run the 5 sub-steps in order. Do not skip ahead — each gate exists\n    because skipping it produced a real class of HYP-backlog noise\n    (audit of HYP-001..026 found ~50% drop+merge waste — the baseline).\n\n    #### 4.5.a Filter — drop noise BEFORE formulating\n\n    Apply each criterion. The candidate survives only if ALL hold:\n\n    - **Recurrence or cost:** observed ≥2 times in this session, OR\n      cost ≥5 minutes wasted, OR has at least one named adjacent\n      flavor (sibling failure mode with the same shape).\n    - **Not already covered:** check the active-HYP set (from the\n      session's evidence block, or `rack context HYP-NNN` per candidate)\n      and verify no active HYP describes the same mechanism. If one does,\n      the finding becomes evidence for that HYP via `ai-hats reflect\n      issue` — but presented as a verdict contribution, not a new candidate.\n    - **Cause is hypothesised, not known:** if you can name the\n      mechanism AND the fix with confidence, it belongs in the task\n      tracker (step 7), not in HYP — HYP is for behavioural patterns\n      whose cause is uncertain enough to need observation across\n      sessions.\n\n    A finding that fails any criterion is **not** a HYP. Either drop\n    it, route it to step 7 (task card), or attach it as evidence to an\n    existing HYP. Single-instance one-off contexts almost always fail\n    the recurrence gate — most of the dropped HYPs in that\n    baseline were single-context (HYP-023, 024, 026).\n\n    #### 4.5.b Root-cause — chain to Step 3, do not restate\n\n    For every surviving candidate, take the 5 Whys output from Step 3\n    and walk it down until the answer is a **behaviour mechanism**,\n    not a behaviour description. Stop when the next \"why?\" would name\n    an environmental \u002F system constraint outside the agent's control.\n\n    Mechanism = \"rule X fires only in context Y because…\",\n                \"agent generalises from prior Z without re-reading…\",\n                \"agent treats each flavor as a one-shot lesson…\"\n    NOT mechanism = \"agent forgot to do X\",\n                    \"agent assumed X\" (assumed *why*? keep asking).\n\n    Sanity check: if your CAUSE sentence is the same content as the\n    OBSERVATION sentence with different wording, you have not reached\n    the mechanism. Go one level deeper. (HYP-023 is the negative\n    example: cause text \"agent assumes transition requires pre-merge\"\n    sounded like a mechanism but was actually surface — the HYP was\n    refuted because the real mechanism was muscle-memory git-merge\n    reflex unrelated to FSM assumptions.)\n\n    #### 4.5.c Generalize — collapse to class\n\n    **Umbrella-first gate — ask this BEFORE you frame anything:**\n    do ≥2 surviving candidates share one root failure mode? If yes,\n    write **ONE** class-level HYP now that names each context as a\n    flavor. Do **not** file N atomic HYPs and leave the supervisor to\n    merge them (the HYP-043\u002F044\u002F045 → HYP-046 churn: three filings, all\n    stalled, replaced by one umbrella, two wasted CLI round-trips).\n\n    Lay all surviving candidates side-by-side. For any two (or more)\n    that share a mechanism, fold them into ONE class-level HYP that\n    names both contexts as flavors. Reference HYP-022 as the model:\n    one HYP, four named flavors of the same shell-quoting class.\n\n    Also re-scan active HYPs for class-level overlap — if your new\n    class-HYP is a generalisation of an existing narrow HYP, propose\n    superseding the narrow one rather than adding a third HYP.\n\n    #### 4.5.d Frame — present with concrete trace, not a one-liner\n\n    For each surviving candidate, write a block in this shape:\n\n    ```\n    [N] OBSERVATION  (class-level: the behavioural pattern, NOT the instance)\n        \u003Cone line naming the pattern the agent systematically exhibits>\n\n        WHAT WENT WRONG (concrete trace, 3-7 lines)\n        - \u003Cwhat the agent did, with specific commands \u002F files \u002F cites>\n        - \u003Cwhat was expected instead>\n        - \u003Chow it surfaced: error message, supervisor pushback, lost time>\n        - \u003Cif class-level: name 2+ flavors observed or named>\n\n        HYPOTHESISED CAUSE  (class-level mechanism)\n        \u003Cmechanism from 4.5.b — one or two sentences>\n\n        SCOPE\n        single | class (list adjacent flavors)\n\n        WHY RECORD\n        \u003Cwhat evidence the observation window collects;\n         what changes if confirmed vs refuted>\n    ```\n\n    **Class-level framing rule.** The OBSERVATION (and the HYP title you\n    derive from it) and the HYPOTHESISED CAUSE must read as a *class* —\n    the pattern and its mechanism — never as the single trace you saw.\n    The concrete instance (this session's commands, files, the one\n    symptom) belongs ONLY in WHAT WENT WRONG. Litmus: if the OBSERVATION\n    line can be rewritten as \"in \u003Csession\u002Fturn> the agent did \u003CX>\", it is\n    still instance-level — lift it one level before presenting.\n\n    - ✗ instance: «агент сообщил \"zero blast radius\" до конца прогона»\n    - ✓ class: «агент формулирует влияние изменения из guess, а не из\n      evidence» — the \"zero blast radius\" trace moves into WHAT WENT WRONG\n\n    One-liner candidate lists are forbidden — the supervisor needs the\n    trace to decide keep\u002Fdrop without re-reading the session log.\n\n    #### 4.5.e Confirm — supervisor decides, then act\n\n    Present the framed blocks. Ask: \"Record as hypotheses? (all \u002F\n    1,3 \u002F none \u002F discuss)\". On confirmation, call `ai-hats reflect\n    issue` (one per item, `--bg` ok). Do NOT call before confirmation\n    — the HYP backlog is the supervisor's source of truth, not an\n    agentic auto-flush.\n\n    Skip 4.5 entirely if no findings survive 4.5.a.\n\n5. **Quantify:** \"7 iterations wasted\", \"3 failed attempts before pivot\" —\n   numbers make the impact visible.\n\n6. **Archive:** Save report to `\u003Cai_hats_dir>\u002Fsessions\u002Fretros\u002FYYYY-MM-DD-retro-\u003Ctitle>.md`.\n\n   Self-retros are personal agent reflections — free-form markdown is fine.\n   They are NOT part of the automated feedback loop (reflect-session handles\n   that via `hats-reflect-session\u002Fv1`).\n\n7. **Backlog:** For deferred improvements, create task cards via `rack create` (skill **hatrack**).\n   Every identified improvement must either be fixed now or tracked.\n\n## Completion\n\n- Retrospective report saved to `\u003Cai_hats_dir>\u002Fsessions\u002Fretros\u002F`\n- Every improvement either applied or tracked as a task card\n- Quantified impact (wasted iterations, failed attempts)\n\n## Anti-Patterns\n\n- Vague findings (\"things could be better\") — be specific with numbers and examples\n- Only listing problems without root cause analysis — use 5 Whys\n- Skipping the backlog step — improvements that aren't tracked will be forgotten\n- Conflating a **fix** (concrete change → task card) with a **pattern**\n  (behavioural tendency → `reflect issue`). If the wording is \"do X\",\n  it's a fix; if it's \"agent does X\", it's a pattern.\n- Calling `ai-hats reflect issue` without the supervisor's confirmation —\n  the hypothesis backlog is their source of truth, not an agentic\n  auto-flush. Always present the candidate list and wait.\n- **Skipping the 4.5.a filter** and dumping every candidate at the\n  supervisor — when the agent retracts after pushback (\"на самом деле\n  это шум\"), the candidate failed the filter and should never have been\n  presented. Run the filter first.\n- **Restating the observation as the cause** in 4.5.b — if your CAUSE\n  sentence paraphrases the OBSERVATION, you stopped one \"why?\" short.\n- **One-liner candidate presentation** — Frame (4.5.d) demands a\n  concrete trace. A line like \"agent omits paths in tracker references\"\n  forces the supervisor to reconstruct context; the trace block makes\n  the keep\u002Fdrop call obvious.\n- **Filing N narrow HYPs that share a mechanism** — collapse to one\n  class HYP at 4.5.c (umbrella-first gate). HYP-022 is the model (one\n  HYP, four flavors); HYP-043\u002F044\u002F045 → HYP-046 is the anti-model.\n- **Instance-level OBSERVATION \u002F CAUSE** — framing the candidate as the\n  single trace you saw (\"agent said 'zero blast radius' before the run\n  finished\") instead of the class (\"agent states change-impact from\n  guess, not evidence\"). The instance belongs in WHAT WENT WRONG; the\n  OBSERVATION and HYPOTHESISED CAUSE must be class-level (4.5.d).\n","# Self-Retrospective\n\nAnalyze completed or failed work to identify systemic improvements.\n\n## When to Use\n\n- When the supervisor explicitly asks to \"write a retro\" \u002F \"do a\n  retrospective\" — follow THIS protocol (4.5 filter, step-7 backlog gates);\n  do not free-form a retro file by pattern-matching an existing one.\n- At the end of a multi-task work session (wrap-up across several closed\n  tasks) — not only single-task completion.\n- After task completion (especially with failures or backtracks)\n- After a failed task (mandatory)\n- When the backlog manager transitions to `review` or `failed` state\n\n**Not** the automated session-review loop. `reflect-session`\n(`hats-reflect-session\u002Fv1`) is a pipeline the harness runs to triage\nhypotheses\u002Fproposals; `self-retrospective` is the personal agent reflection\nyou run by hand (free-form markdown, archived under `sessions\u002Fretros\u002F`). And\n**not** a factual outcome record — that is **task-summary**; this skill is the\n*improvement* analysis (why the work went as it did, what to change next time).\n\n## Procedure\n\n1. **Facts:** Chronologically list all errors, backtracks, wasted iterations,\n   or unexpected findings from the session.\n\n2. **Classify:** Assign each issue to a category:\n   - **Knowledge** — missing context or incorrect assumption about the system\n   - **Environment** — tooling, permissions, infrastructure issues\n   - **Process** — skipped steps, wrong order, missing verification\n   - **Communication** — misunderstood requirements, ambiguous instructions\n   - **Assumption** — untested belief that turned out wrong\n\n3. **Root Cause (5 Whys):** For each significant issue, ask \"Why?\" iteratively\n   until you reach an actionable root cause. Stop at the level where a fix\n   is practical.\n\n4. **Improvements:** Propose specific, concrete changes:\n   - Rule or skill updates (with exact content)\n   - Workflow changes (with steps)\n   - New checks or verification steps\n     Focus on **systemic** fixes over one-off patches.\n\n4.5 **Hypothesis candidates:** Surface improvements that describe a\n**behavioural pattern** (the agent systematically does X) — those go\ninto the hypothesis backlog, not the task tracker. The supervisor\nchooses what to record.\n\n    Run the 5 sub-steps in order. Do not skip ahead — each gate exists\n    because skipping it produced a real class of HYP-backlog noise\n    (audit of HYP-001..026 found ~50% drop+merge waste — the baseline).\n\n    #### 4.5.a Filter — drop noise BEFORE formulating\n\n    Apply each criterion. The candidate survives only if ALL hold:\n\n    - **Recurrence or cost:** observed ≥2 times in this session, OR\n      cost ≥5 minutes wasted, OR has at least one named adjacent\n      flavor (sibling failure mode with the same shape).\n    - **Not already covered:** check the active-HYP set (from the\n      session's evidence block, or `rack context HYP-NNN` per candidate)\n      and verify no active HYP describes the same mechanism. If one does,\n      the finding becomes evidence for that HYP via `ai-hats reflect\n      issue` — but presented as a verdict contribution, not a new candidate.\n    - **Cause is hypothesised, not known:** if you can name the\n      mechanism AND the fix with confidence, it belongs in the task\n      tracker (step 7), not in HYP — HYP is for behavioural patterns\n      whose cause is uncertain enough to need observation across\n      sessions.\n\n    A finding that fails any criterion is **not** a HYP. Either drop\n    it, route it to step 7 (task card), or attach it as evidence to an\n    existing HYP. Single-instance one-off contexts almost always fail\n    the recurrence gate — most of the dropped HYPs in that\n    baseline were single-context (HYP-023, 024, 026).\n\n    #### 4.5.b Root-cause — chain to Step 3, do not restate\n\n    For every surviving candidate, take the 5 Whys output from Step 3\n    and walk it down until the answer is a **behaviour mechanism**,\n    not a behaviour description. Stop when the next \"why?\" would name\n    an environmental \u002F system constraint outside the agent's control.\n\n    Mechanism = \"rule X fires only in context Y because…\",\n                \"agent generalises from prior Z without re-reading…\",\n                \"agent treats each flavor as a one-shot lesson…\"\n    NOT mechanism = \"agent forgot to do X\",\n                    \"agent assumed X\" (assumed *why*? keep asking).\n\n    Sanity check: if your CAUSE sentence is the same content as the\n    OBSERVATION sentence with different wording, you have not reached\n    the mechanism. Go one level deeper. (HYP-023 is the negative\n    example: cause text \"agent assumes transition requires pre-merge\"\n    sounded like a mechanism but was actually surface — the HYP was\n    refuted because the real mechanism was muscle-memory git-merge\n    reflex unrelated to FSM assumptions.)\n\n    #### 4.5.c Generalize — collapse to class\n\n    **Umbrella-first gate — ask this BEFORE you frame anything:**\n    do ≥2 surviving candidates share one root failure mode? If yes,\n    write **ONE** class-level HYP now that names each context as a\n    flavor. Do **not** file N atomic HYPs and leave the supervisor to\n    merge them (the HYP-043\u002F044\u002F045 → HYP-046 churn: three filings, all\n    stalled, replaced by one umbrella, two wasted CLI round-trips).\n\n    Lay all surviving candidates side-by-side. For any two (or more)\n    that share a mechanism, fold them into ONE class-level HYP that\n    names both contexts as flavors. Reference HYP-022 as the model:\n    one HYP, four named flavors of the same shell-quoting class.\n\n    Also re-scan active HYPs for class-level overlap — if your new\n    class-HYP is a generalisation of an existing narrow HYP, propose\n    superseding the narrow one rather than adding a third HYP.\n\n    #### 4.5.d Frame — present with concrete trace, not a one-liner\n\n    For each surviving candidate, write a block in this shape:\n\n    ```\n    [N] OBSERVATION  (class-level: the behavioural pattern, NOT the instance)\n        \u003Cone line naming the pattern the agent systematically exhibits>\n\n        WHAT WENT WRONG (concrete trace, 3-7 lines)\n        - \u003Cwhat the agent did, with specific commands \u002F files \u002F cites>\n        - \u003Cwhat was expected instead>\n        - \u003Chow it surfaced: error message, supervisor pushback, lost time>\n        - \u003Cif class-level: name 2+ flavors observed or named>\n\n        HYPOTHESISED CAUSE  (class-level mechanism)\n        \u003Cmechanism from 4.5.b — one or two sentences>\n\n        SCOPE\n        single | class (list adjacent flavors)\n\n        WHY RECORD\n        \u003Cwhat evidence the observation window collects;\n         what changes if confirmed vs refuted>\n    ```\n\n    **Class-level framing rule.** The OBSERVATION (and the HYP title you\n    derive from it) and the HYPOTHESISED CAUSE must read as a *class* —\n    the pattern and its mechanism — never as the single trace you saw.\n    The concrete instance (this session's commands, files, the one\n    symptom) belongs ONLY in WHAT WENT WRONG. Litmus: if the OBSERVATION\n    line can be rewritten as \"in \u003Csession\u002Fturn> the agent did \u003CX>\", it is\n    still instance-level — lift it one level before presenting.\n\n    - ✗ instance: «агент сообщил \"zero blast radius\" до конца прогона»\n    - ✓ class: «агент формулирует влияние изменения из guess, а не из\n      evidence» — the \"zero blast radius\" trace moves into WHAT WENT WRONG\n\n    One-liner candidate lists are forbidden — the supervisor needs the\n    trace to decide keep\u002Fdrop without re-reading the session log.\n\n    #### 4.5.e Confirm — supervisor decides, then act\n\n    Present the framed blocks. Ask: \"Record as hypotheses? (all \u002F\n    1,3 \u002F none \u002F discuss)\". On confirmation, call `ai-hats reflect\n    issue` (one per item, `--bg` ok). Do NOT call before confirmation\n    — the HYP backlog is the supervisor's source of truth, not an\n    agentic auto-flush.\n\n    Skip 4.5 entirely if no findings survive 4.5.a.\n\n5. **Quantify:** \"7 iterations wasted\", \"3 failed attempts before pivot\" —\n   numbers make the impact visible.\n\n6. **Archive:** Save report to `\u003Cai_hats_dir>\u002Fsessions\u002Fretros\u002FYYYY-MM-DD-retro-\u003Ctitle>.md`.\n\n   Self-retros are personal agent reflections — free-form markdown is fine.\n   They are NOT part of the automated feedback loop (reflect-session handles\n   that via `hats-reflect-session\u002Fv1`).\n\n7. **Backlog:** For deferred improvements, create task cards via `rack create` (skill **hatrack**).\n   Every identified improvement must either be fixed now or tracked.\n\n## Completion\n\n- Retrospective report saved to `\u003Cai_hats_dir>\u002Fsessions\u002Fretros\u002F`\n- Every improvement either applied or tracked as a task card\n- Quantified impact (wasted iterations, failed attempts)\n\n## Anti-Patterns\n\n- Vague findings (\"things could be better\") — be specific with numbers and examples\n- Only listing problems without root cause analysis — use 5 Whys\n- Skipping the backlog step — improvements that aren't tracked will be forgotten\n- Conflating a **fix** (concrete change → task card) with a **pattern**\n  (behavioural tendency → `reflect issue`). If the wording is \"do X\",\n  it's a fix; if it's \"agent does X\", it's a pattern.\n- Calling `ai-hats reflect issue` without the supervisor's confirmation —\n  the hypothesis backlog is their source of truth, not an agentic\n  auto-flush. Always present the candidate list and wait.\n- **Skipping the 4.5.a filter** and dumping every candidate at the\n  supervisor — when the agent retracts after pushback (\"на самом деле\n  это шум\"), the candidate failed the filter and should never have been\n  presented. Run the filter first.\n- **Restating the observation as the cause** in 4.5.b — if your CAUSE\n  sentence paraphrases the OBSERVATION, you stopped one \"why?\" short.\n- **One-liner candidate presentation** — Frame (4.5.d) demands a\n  concrete trace. A line like \"agent omits paths in tracker references\"\n  forces the supervisor to reconstruct context; the trace block makes\n  the keep\u002Fdrop call obvious.\n- **Filing N narrow HYPs that share a mechanism** — collapse to one\n  class HYP at 4.5.c (umbrella-first gate). HYP-022 is the model (one\n  HYP, four flavors); HYP-043\u002F044\u002F045 → HYP-046 is the anti-model.\n- **Instance-level OBSERVATION \u002F CAUSE** — framing the candidate as the\n  single trace you saw (\"agent said 'zero blast radius' before the run\n  finished\") instead of the class (\"agent states change-impact from\n  guess, not evidence\"). The instance belongs in WHAT WENT WRONG; the\n  OBSERVATION and HYPOTHESISED CAUSE must be class-level (4.5.d).\n",{"name":5,"ai_hats":39,"license":15,"description":7},{"type":40,"schema_version":41},"skill",1,"master","f922a7f627276366873d546942d9ad899a86f71a","packages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fcore\u002Fskills\u002Fself-retrospective","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats","https:\u002F\u002Fgithub.com\u002Fmuratovv","2026-11-05 20:59:10",{"source":8,"skillName":5,"commands":49,"prompt":59,"zipAvailable":60},[50,53,56],{"tool":51,"command":52},"npx","npx skills add muratovv\u002Fai-hats --skill self-retrospective",{"tool":54,"command":55},"bunx","bunx skills add muratovv\u002Fai-hats --skill self-retrospective",{"tool":57,"command":58},"pnpm","pnpm dlx skills add muratovv\u002Fai-hats --skill self-retrospective","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fcore\u002Fskills\u002Fself-retrospective\u002FSKILL.md",true,[],[63,87,110,133,156,170],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":34},"01M6PVFB1SM738R2SRSRSF7CY6","tool-evaluation-protocol","Time-bounded protocol for evaluating new tools, libraries, or methodologies before adoption. Use when evaluating a candidate dependency for the project, researching a methodology (e.g. ML\u002FLLM frameworks, testing approaches), or following up on a hype-driven recommendation.","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fusage\u002Fskills\u002Ftool-evaluation-protocol\u002FSKILL.md",[],[70,71,72,73,74,75,76,77,78,79,27,80,28,81,82,83,84,32,85,86],"protocol","Evaluation","tool","tool evaluation protocol","testing","ml","libraries","adoption","methodology","Tools","dependency","frameworks","time","project","recommendation","bounded","candidate",{"skillNo":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":34},"01M6PVF7HATBEX1G7N50E2QT57","systematic-debugging","4-phase bug-fix protocol (evidence, pattern, hypothesis, verify). Use when investigating any error, failure, or unexpected behavior, handling bug reports from users or monitoring, or diagnosing test failures with non-obvious causes.","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fusage\u002Fskills\u002Fsystematic-debugging\u002FSKILL.md",[],[94,95,96,97,70,98,99,100,101,25,27,102,103,104,105,106,107,31,108,109],"systematic debugging","debugging","systematic","test","monitoring","hypothesis","verify","behavior","evidence","bug","reports","pattern","users","error","bug-fix","phase",{"skillNo":111,"name":112,"nameZh":6,"description":113,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":114,"categories":115,"keywords":121,"publishedAt":34},"01M6PVF3MTQN3RPF73AN34856J","skill-template","Canonical template and validation guide for ai-hats skills. Use when creating a new skill, reviewing an existing one for structural compliance, or deciding which pattern fits a new behavior.","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fusage\u002Fskills\u002Fskill-template\u002FSKILL.md",[116],{"categoryNo":117,"categoryName":118,"taxonomyType":119,"taxonomyTypeDesc":120},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[122,123,124,125,126,101,127,27,128,105,129,130,131,32,132],"template","skill template","validation","compliance","hats","guide","new skill","structural","canonical","which","Existing",{"skillNo":134,"name":135,"nameZh":6,"description":136,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":137,"categories":138,"keywords":140,"publishedAt":34},"01M6PVF06SCV28QARM5MQPANC9","skill-optimization","Audit and refactor library components to eliminate redundancy and staleness. Use during periodic library maintenance, after adding multiple new skills or rules, when a role's token budget is too high, or after retrospectives identify component issues.","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fusage\u002Fskills\u002Fskill-optimization\u002FSKILL.md",[139],{"categoryNo":117,"categoryName":118,"taxonomyType":119,"taxonomyTypeDesc":120},[141,142,143,144,145,146,22,147,148,149,27,150,151,152,153,154,155],"Optimization","skill optimization","components","component","issues","library","refactor","Budget","Rules","periodic","maintenance","token","token budget","role","high",{"skillNo":157,"name":158,"nameZh":6,"description":159,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":160,"categories":161,"keywords":162,"publishedAt":34},"01M6PVEXM9717016T9MW82HDCT","silent-fallback-gate","PreToolUse hook — refuses an Edit or Write that adds an inert broad except. Use when a Python edit was blocked for a silent fallback.","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fusage\u002Fskills\u002Fsilent-fallback-gate\u002FSKILL.md",[],[163,164,165,166,167,168,27,169,32],"gate","fallback","silent fallback gate","python","write","edit","hook",{"skillNo":171,"name":172,"nameZh":6,"description":173,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":174,"categories":175,"keywords":177,"publishedAt":34},"01M6PVESM49C5NT7SB7N3F76VS","retro-to-framework","Convert an observation about agent behavior — from a retrospective or from a single incident — into a framework-level change (hook, rule, skill, skill update). Use after a retrospective identifies problems that are not project-specific, after an incident where an agent misapplied a component it had read, when CLAUDE.md band-aids accumulate (more than 3 per-project rules that could be generic), or when the same problem recurs across multiple projects.","https:\u002F\u002Fgithub.com\u002Fmuratovv\u002Fai-hats\u002Fblob\u002Fmaster\u002Fpackages\u002Fai-hats-library\u002Fsrc\u002Fai_hats_library\u002Fexperimental\u002Fusage\u002Fskills\u002Fretro-to-framework\u002FSKILL.md",[176],{"categoryNo":117,"categoryName":118,"taxonomyType":119,"taxonomyTypeDesc":120},[178,23,179,180,144,19,181,182,149,101,27,183,184,83,185,186,187,188,189,190],"framework","retro to framework","change","projects","convert","incident","across","problem","about","more","skill update","single","generic"]