[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6CHT5VCMGE14XTGVN9E08KA":3,"skill-same:01M6CHT5VCMGE14XTGVN9E08KA":97},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":22,"publishedAt":41,"linkOnly":42,"markdownContent":43,"markdownBody":44,"frontMatter":45,"refName":60,"commitSha":61,"skillDirectory":62,"forksCount":63,"repositoryHtmlUrl":64,"repositoryHomepageUrl":65,"ownerHtmlUrl":66,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":67,"install":68,"resources":81},"01M6CHT5VCMGE14XTGVN9E08KA","retro",null,"[ADD v{{VERSION}}] Run a retrospective — context-aware, data-driven review with pre-populated tables","MountainUnicorn\u002Fadd","MountainUnicorn","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F199411549?v=4","01M6CHRAWFWD275ANP1S2A0HXC","01M6CHRB0QE1678D913B438SAR",11,"MIT","https:\u002F\u002Fgithub.com\u002FMountainUnicorn\u002Fadd\u002Fblob\u002Fmain\u002Fcore\u002Fskills\u002Fretro\u002FSKILL.md",[17],{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[5,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40],"Rules","summary","reference","version","session","context","edit","write","read","grep","review","learning","Bash","data","dd","run","retrospective","telemetry","2026-11-10 10:12:30",false,"---\ndescription: \"[ADD v{{VERSION}}] Run a retrospective — context-aware, data-driven review with pre-populated tables\"\nargument-hint: \"[--agent-summary] [--since YYYY-MM-DD] [--scope feature|sprint|session] [--dry-run]\"\nallowed-tools: [Read, Write, Edit, Glob, Grep, Bash, AskUserQuestion, TodoWrite]\nreferences: [\"learning-reference.md\", \"maturity-matrix.md\", \"skill-epilogue.md\", \"rules\u002Ftelemetry.md\"]\n---\n\n# ADD Retro Command v{{VERSION}}\n\nContext-aware retrospective that auto-gathers data, classifies human directives and agent observations into scoped tables, and presents pre-populated findings for the human to refine — not recall from scratch.\n\nTwo modes:\n- **Interactive retro** (default) — Data-driven review with pre-populated tables\n- **Agent summary** (`--agent-summary`) — Quick non-interactive observations review\n\n## Pre-Flight\n\n1. Read `.add\u002Fconfig.json` for project context (name, maturity, stack)\n2. Read all 3 knowledge tiers:\n   a. **Tier 1:** Read `${CLAUDE_PLUGIN_ROOT}\u002Fknowledge\u002Fglobal.md`\n   b. **Tier 2:** Read `~\u002F.claude\u002Fadd\u002Flibrary.json` (fall back to `library.md`)\n   c. **Tier 3:** Read `.add\u002Flearnings.json` (fall back to `learnings.md`)\n3. Read `~\u002F.claude\u002Fadd\u002Fprofile.md` if it exists\n4. Determine the retro window:\n   - If `--since` provided, use that date\n   - If previous retro exists in `.add\u002Fretros\u002F`, use that date as start\n   - Otherwise, use project creation date from config\n5. Check for session handoff — per the Session-Handoff Preflight in `${CLAUDE_PLUGIN_ROOT}\u002Freferences\u002Fskill-epilogue.md`\n\n---\n\n## Mode 1: Interactive Retro (default)\n\n### Phase 1: Detect Session Context\n\nDetermine how the human spent the retro window:\n\n1. **Read `.add\u002Faway-logs\u002F`** — count away sessions, total autonomous hours\n2. **Read git log** — count commits, branches, PRs merged\n3. **Read `.add\u002Fhandoff.md`** — check for interactive session indicators\n4. **Classify context:**\n   - **Autonomous:** >70% of retro window was away-mode sessions\n   - **Collaborative:** \u003C30% away-mode, active interactive exchanges\n   - **Mixed:** between 30-70% away-mode\n\nThe context classification affects the flow:\n- **Autonomous:** Human has less context → skip \"what went well\", reduce questions\n- **Collaborative\u002FMixed:** Full question set\n\n### Phase 2: Auto-Gather Metrics\n\nCollect data for the retro window from all sources:\n\n- **Git log:** commit count, branches merged, PRs\n- **Specs:** count completed (status: Complete) in `specs\u002F`\n- **Learnings:** count new entries in `.add\u002Flearnings.json` since last retro (by date field)\n- **Workstation learnings:** count new entries in `~\u002F.claude\u002Fadd\u002Flibrary.json` since last retro\n- **Observations:** read `.add\u002Fobservations.md` entries since last retro\n- **Away logs:** count sessions, total duration from `.add\u002Faway-logs\u002F`\n- **Handoff:** current state from `.add\u002Fhandoff.md`\n\nPresent the summary:\n\n```\nRETROSPECTIVE — {PROJECT_NAME}\nPeriod: {start_date} → {today}\nContext: {Collaborative|Autonomous|Mixed} ({detail})\n\nDuring this period:\n  Specs completed: {N}\n  Commits: {N}\n  Agent learnings recorded: {N} ({N} project, {N} workstation)\n  Human directives captured: {N}\n  Away-mode sessions: {N}\n```\n\n### Phase 3: Human Directives (Table 1)\n\nExtract directives the human gave during the retro window from:\n- **Handoff files** — decisions, explicit instructions\n- **Learnings entries** with `classified_by: \"human\"` — human-reclassified knowledge\n- **`.add\u002Fobservations.md`** — entries recording human feedback or directives\n- **Conversation context** — explicit instructions from the current or recent sessions\n\nFor each directive, **classify scope:**\n\n| Scope | Signal |\n|-------|--------|\n| project | References specific files, config, routes, schemas unique to this project |\n| workstation | References tools, libraries, workflows that apply across projects |\n| organization | References team, org, or company patterns (stub — accept but note \"future tier\") |\n| community | References universal methodology insights (stub — accept but note \"future tier\") |\n\n**If no directives found:** Skip Table 1 entirely, move to Phase 4. Do not show an empty table.\n\n**If directives found:** Present:\n\n```\nIn our recent sessions, you provided some key insights:\n\n━━━ TABLE 1: YOUR DIRECTIVES ━━━\n| # | Directive | Scope | Source |\n|---|-----------|-------|--------|\n| 1 | {directive text} | {scope} | {source} |\n| 2 | {directive text} | {scope} | {source} |\n\nAre these captured correctly in learnings? Any to add, remove, or reclassify?\n```\n\nWait for human confirmation. Apply any changes to learnings JSON files.\n\n### Phase 4: Agent Observations (Tables 2 & 3)\n\nRead agent-generated entries since last retro:\n- **Table 2 (Project):** Entries from `.add\u002Flearnings.json` where `classified_by: \"agent\"` and `date` within retro window\n- **Table 3 (Workstation):** Entries from `~\u002F.claude\u002Fadd\u002Flibrary.json` where `classified_by: \"agent\"` and `date` within retro window\n\n**Skip any table that has zero entries.** Do not show empty tables.\n\n**ADD Methodology Adherence Self-Assessment:**\n\nInclude in the agent observations a self-assessment of how well ADD methodology was followed during the retro window. Check each rule:\n\n| Rule | How to Assess |\n|------|---------------|\n| Spec-before-code | Were any features implemented without a spec? Check git log for implementation commits vs spec dates |\n| TDD cycles | Were tests written before implementation? (N\u002FA for markdown-only plugins) |\n| Auto-handoffs | Were handoffs written after commits and major work? Check `.add\u002Fhandoff.md` timestamps vs git log |\n| Learning checkpoints | Were learnings recorded at trigger points (post-verify, post-tdd, post-deploy, post-away)? Count expected vs actual |\n| Quality gates | Were `\u002Fadd:verify` runs done? Did they pass first time? |\n| Source control | Were conventional commits used? Check last N commits for pattern compliance |\n\nFormat methodology adherence as a **table** (not inline checkmarks — tables are easier to scan):\n\n```\n━━━ TABLE 2: AGENT OBSERVATIONS (PROJECT) ━━━\n| # | Observation | Severity |\n|---|-------------|----------|\n| 1 | {observation} | {severity} |\n\nADD Methodology Adherence:\n| Rule | Status | Detail |\n|------|--------|--------|\n| Spec-before-code | ✓ Pass | {assessment} |\n| Auto-handoffs | ✓ Pass | {assessment} |\n| Learning checkpoints | ✗ Fail | {assessment with specifics} |\n```\n\n**Do NOT show Table 3 if it has zero entries.** Skip it silently — no heading, no empty table, no mention.\n\nThen ask:\n\n```\nHelp me polish these. Do you disagree or wish to modify any of these learnings?\n```\n\nWait for human input. If the human identifies data gaps (missing checkpoints, incorrect entries), fix them **in the background** — queue the writes and continue the retro flow immediately. Do not block the human while writing JSON entries or regenerating markdown views.\n\n### Phase 5: Targeted Questions\n\nAsk one at a time. Adapt based on session context:\n\n**Q1 (skip if Autonomous context):** \"What went well?\"\n→ Only ask if the human had enough interactive context to answer meaningfully.\n\n**Q2:** \"What needed improvement that was not included already in our learnings?\"\n→ Scoped to gaps — the tables already surfaced known issues.\n\n**Q3 (rate-limited — 1x per calendar day):** \"On a scale of 0.0 to 9.0, how well are we working together?\"\n→ To check rate limit: read `.add\u002Fretros\u002F` for any file matching `retro-{today's date}*.md` that already has a `Human Collaboration` score populated. If found, skip this question.\n\n**Q4 (rate-limited — 1x per calendar day):** \"Any improvements for Agent Driven Development you would suggest?\"\n→ Same rate-limit check as Q3. These two are always asked\u002Fskipped together.\n\n### Phase 6: Agent Self-Assessment Scores\n\nThe agent provides two self-assessed scores with evidence. These are NOT asked of the human — the agent generates them from data.\n\n**ADD Methodology Effectiveness (0.0-9.0):**\n- Based on methodology adherence from Phase 4\n- Consider: spec coverage, TDD compliance, handoff discipline, checkpoint completeness, quality gate usage\n- Must cite specific evidence (not vague claims)\n\n**Swarm Effectiveness (0.0-9.0):**\n- How well did agents collectively build together?\n- Consider: parallel subagent usage, context handoff quality between sessions, duplicate work avoided, session continuity\n- Must cite specific evidence\n\nFormat:\n\n```\n━━━ AGENT SELF-ASSESSMENT ━━━\nADD Methodology Effectiveness: {X.X} \u002F 9.0\n  Evidence: {specific evidence from the retro window}\n\nSwarm Effectiveness: {X.X} \u002F 9.0\n  Evidence: {specific evidence from the retro window}\n```\n\nScores must be justified by evidence. If the agent detects its own score seems inflated relative to evidence, adjust downward. Honesty over optimism.\n\n### Phase 7: Record and Update\n\n1. **Write retro archive:**\n   Use `${CLAUDE_PLUGIN_ROOT}\u002Ftemplates\u002Fretro.md.template` as structure.\n   Write to `.add\u002Fretros\u002Fretro-{date}.md` (create directory if needed).\n   Fill all sections with data from the retro.\n\n2. **Store scores:**\n   Read `.add\u002Fretro-scores.json` (create from `${CLAUDE_PLUGIN_ROOT}\u002Ftemplates\u002Fretro-scores.json.template` if doesn't exist).\n   Append entry:\n   ```json\n   {\n     \"date\": \"{YYYY-MM-DD}\",\n     \"collab_score\": {X.X},\n     \"add_effectiveness\": {X.X},\n     \"swarm_effectiveness\": {X.X},\n     \"retro_file\": \".add\u002Fretros\u002Fretro-{date}.md\",\n     \"context\": \"{collaborative|autonomous|mixed}\"\n   }\n   ```\n   If collab score was rate-limited (skipped), use `null` for `collab_score`.\n\n3. **Store ADD feedback:**\n   If the human provided ADD improvement suggestions, append to `.add\u002Fadd-feedback.md`:\n   ```markdown\n   ## {YYYY-MM-DD}\n   - **Suggestion:** {text}\n   - **Retro:** .add\u002Fretros\u002Fretro-{date}.md\n   - **Streamed:** false\n   ```\n   Create the file with `# ADD Methodology Feedback` header if it doesn't exist.\n\n4. **Update learnings:**\n   Write new learning entries from the retro as structured JSON to the appropriate file (`.add\u002Flearnings.json` or `~\u002F.claude\u002Fadd\u002Flibrary.json`). Follow the checkpoint process in `${CLAUDE_PLUGIN_ROOT}\u002Freferences\u002Flearning-reference.md`:\n   - Classify scope for each new learning\n   - Write to the appropriate JSON file\n   - Regenerate the corresponding markdown view\n\n5. **Scope review and reclassification:**\n   Review entries classified by agents since the last retro:\n   ```\n   SCOPE REVIEW — Agent-classified entries since last retro:\n     L-{NNN}: \"{title}\" — classified as {scope} by agent\n     WL-{NNN}: \"{title}\" — classified as {scope} by agent\n\n   Any reclassifications needed?\n   ```\n   For each reclassification the human approves:\n   - Move the entry between JSON files\n   - Update `scope` field and `classified_by` to `\"human\"`\n   - Assign new ID appropriate to target file\n   - Regenerate both markdown views\n\n   **Auto-promotion proposals (batch):**\n\n   In addition to asking per-entry, scan `.add\u002Flearnings.json` entries since the last retro and auto-propose **workstation promotion candidates** — entries that look universal enough to belong in `~\u002F.claude\u002Fadd\u002Flibrary.json` for cross-project use.\n\n   Promotion signals (ANY of):\n   - `severity >= \"high\"` AND `category = \"anti-pattern\"`\n   - Body contains \"always\", \"never\", \"every time\", \"on every\", \"default\" — language indicating a general rule rather than a project-specific fact\n   - `stack` entries describe tooling that the user's other projects in `~\u002F.claude\u002Fadd\u002Fprojects\u002F` also use (check registry stacks)\n   - Body references generic cloud\u002Finfra primitives (GCE, AWS, GKE, EKS, ArgoCD, cert-manager, etc.) without project-specific identifiers\n\n   Present the candidates as one batch table:\n\n   ```\n   PROMOTION CANDIDATES — cross-project workstation learnings\n\n   | # | Entry | Signal | Promote? |\n   |---|-------|--------|----------|\n   | 1 | L-{NNN}: \"{title}\" | {signal that matched} | [y\u002Fn] |\n   | 2 | L-{NNN}: \"{title}\" | {signal that matched} | [y\u002Fn] |\n   ...\n\n   Reply with the numbers to promote (e.g., \"1,3,5\" or \"all\" or \"none\").\n   ```\n\n   For each approved promotion:\n   - Copy entry to `~\u002F.claude\u002Fadd\u002Flibrary.json` with a new `WL-{NNN}` ID\n   - Set `source` to the project name\n   - Set `classified_by` to `\"human\"`\n   - Leave original entry in `.add\u002Flearnings.json` (projects may still reference it)\n   - Regenerate both markdown views at the end\n\n   Why this exists: the agentVoice retro (2026-04-12) surfaced 8 clearly-universal lessons (ArgoCD, CI-disabled, Cloud Build, WebSocket timeout, E2E skip ban) that sat in project markdown for 40 days without promotion. A batch proposal at retro time closes that gap.\n\n6. **Update cross-project persistence:**\n   a. **Profile updates** (`~\u002F.claude\u002Fadd\u002Fprofile.md`): If retro reveals preferences that carry to other projects, ask: \"Add to your ADD profile?\"\n   b. **Project index** (`~\u002F.claude\u002Fadd\u002Fprojects\u002F{name}.json`): Update `last_retro` date and `learnings_count`.\n\n7. **Promote to plugin-global (ADD dev project only):**\n   If running inside the ADD plugin project (detected by `knowledge\u002Fglobal.md` existing as a local file), present Tier 1 promotion candidates.\n\n8. **Apply config\u002Ftemplate changes:** If agreed changes affect the process, make edits now.\n\n9. **Deduplicate knowledge stores:**\n   Check all stores for duplicates or misplaced entries. Report: \"{N} duplicates consolidated, {N} entries relocated.\"\n\n10. **Prune stale entries:**\n    - Observations >30 days old without `[synthesized M-{NNN}]` → archive\n    - Learnings >90 days old without references → flag for human review (never auto-delete)\n\n11. **Regenerate markdown views** after all JSON modifications.\n\n### Phase 8: Observation Synthesis\n\nRead `.add\u002Fobservations.md` and synthesize:\n\n1. Group observations by operation type\n2. Identify patterns (3+ similar = a pattern)\n3. For each pattern, propose a process mutation:\n   ```\n   ### Proposed Mutation: {title}\n   **Skill:** \u002Fadd:{skill-name}\n   **Change:** {concrete change to the skill}\n   **Evidence:** {observation timestamps and summaries}\n   **Expected outcome:** {what should improve}\n   ```\n4. Present proposals to human for approval. Only apply approved mutations.\n\n### Phase 9: Apply Approved Mutations\n\nFor each human-approved mutation:\n1. Read the target skill's SKILL.md\n2. Apply the change\n3. Log in `.add\u002Fmutations.md`\n4. Mark source observations as synthesized\n\n### Phase 10: Process Health Assessment\n\nReview `.add\u002Fmutations.md` for previously applied mutations:\n1. Did the problem recur? → Strengthen mutation\n2. Did the problem stop? → Note positive outcome\n3. New side effects? → Adjust mutation\n4. Report process health summary\n\n### Phase 11: Maturity Promotion Assessment\n\nIf the retro surfaces a promotion request, run an evidence-based check. See maturity promotion criteria:\n\n**POC → Alpha:** At least 3 evidence items (specs, tests, CI, commits, etc.) + core concept validated\n**Alpha → Beta:** Feature specs exist, coverage >50%, CI\u002FCD configured, PR workflow, 2+ environments, conventional commits, TDD evidence\n**Beta → GA:** Coverage >80%, protected branches, release tags, 3+ environments, all gates blocking, 30+ days stability, SLAs defined\n\nDo NOT promote without evidence. Promotion requires proof, not aspiration.\n\n---\n\n## Mode 2: Agent Summary (`--agent-summary`)\n\nQuick, non-interactive mode. Present accumulated observations without a full retro.\n\n1. Read `.add\u002Flearnings.json` and `~\u002F.claude\u002Fadd\u002Flibrary.json`\n2. Filter entries since last retro\n3. Present structured summary:\n\n```\nAGENT OBSERVATIONS — since {last_retro_date}\n\nTECHNICAL DISCOVERIES ({N} entries):\n  - {discovery with date}\n\nARCHITECTURE DECISIONS ({N} entries):\n  - {decision with rationale}\n\nADD METHODOLOGY ADHERENCE:\n  {self-assessment checklist}\n\nWHAT WORKED:\n  - {positive pattern}\n\nWHAT DIDN'T:\n  - {negative pattern}\n\nSUGGESTED CHANGES:\n  1. {suggestion with rationale}\n\nThese are observations only — no changes applied.\nRun \u002Fadd:retro for a full interactive retrospective.\n```\n\n---\n\n## Retro Frequency Guidance\n\nDisplay during the first retro:\n\n```\nRECOMMENDED RETRO CADENCE:\n  - Agent auto-checkpoints: Continuous (after each verify, cycle, away session)\n  - Agent summary (\u002Fadd:retro --agent-summary): Weekly or before planning\n  - Full interactive retro (\u002Fadd:retro): After each feature\u002Fsprint completion,\n    or every 2 weeks — whichever comes first\n```\n\n---\n\n## Score Semantics\n\nAll scores use 0.0-9.0 scale with 1 decimal precision:\n\n| Range | Meaning |\n|-------|---------|\n| 0.0-2.0 | Poor — process not working, significant friction |\n| 2.1-4.0 | Below average — notable gaps, frequent workarounds |\n| 4.1-6.0 | Adequate — functional but room for improvement |\n| 6.1-8.0 | Good — effective with minor issues |\n| 8.1-9.0 | Excellent — highly effective, minimal friction |\n\n---\n\nEnd-of-skill epilogue: follow `${CLAUDE_PLUGIN_ROOT}\u002Freferences\u002Fskill-epilogue.md` (observation + learning checkpoint + progress tracking). The retro's Phase 7 already records its learnings — do not duplicate them; the epilogue's observation line still applies.\n","# ADD Retro Command v{{VERSION}}\n\nContext-aware retrospective that auto-gathers data, classifies human directives and agent observations into scoped tables, and presents pre-populated findings for the human to refine — not recall from scratch.\n\nTwo modes:\n- **Interactive retro** (default) — Data-driven review with pre-populated tables\n- **Agent summary** (`--agent-summary`) — Quick non-interactive observations review\n\n## Pre-Flight\n\n1. Read `.add\u002Fconfig.json` for project context (name, maturity, stack)\n2. Read all 3 knowledge tiers:\n   a. **Tier 1:** Read `${CLAUDE_PLUGIN_ROOT}\u002Fknowledge\u002Fglobal.md`\n   b. **Tier 2:** Read `~\u002F.claude\u002Fadd\u002Flibrary.json` (fall back to `library.md`)\n   c. **Tier 3:** Read `.add\u002Flearnings.json` (fall back to `learnings.md`)\n3. Read `~\u002F.claude\u002Fadd\u002Fprofile.md` if it exists\n4. Determine the retro window:\n   - If `--since` provided, use that date\n   - If previous retro exists in `.add\u002Fretros\u002F`, use that date as start\n   - Otherwise, use project creation date from config\n5. Check for session handoff — per the Session-Handoff Preflight in `${CLAUDE_PLUGIN_ROOT}\u002Freferences\u002Fskill-epilogue.md`\n\n---\n\n## Mode 1: Interactive Retro (default)\n\n### Phase 1: Detect Session Context\n\nDetermine how the human spent the retro window:\n\n1. **Read `.add\u002Faway-logs\u002F`** — count away sessions, total autonomous hours\n2. **Read git log** — count commits, branches, PRs merged\n3. **Read `.add\u002Fhandoff.md`** — check for interactive session indicators\n4. **Classify context:**\n   - **Autonomous:** >70% of retro window was away-mode sessions\n   - **Collaborative:** \u003C30% away-mode, active interactive exchanges\n   - **Mixed:** between 30-70% away-mode\n\nThe context classification affects the flow:\n- **Autonomous:** Human has less context → skip \"what went well\", reduce questions\n- **Collaborative\u002FMixed:** Full question set\n\n### Phase 2: Auto-Gather Metrics\n\nCollect data for the retro window from all sources:\n\n- **Git log:** commit count, branches merged, PRs\n- **Specs:** count completed (status: Complete) in `specs\u002F`\n- **Learnings:** count new entries in `.add\u002Flearnings.json` since last retro (by date field)\n- **Workstation learnings:** count new entries in `~\u002F.claude\u002Fadd\u002Flibrary.json` since last retro\n- **Observations:** read `.add\u002Fobservations.md` entries since last retro\n- **Away logs:** count sessions, total duration from `.add\u002Faway-logs\u002F`\n- **Handoff:** current state from `.add\u002Fhandoff.md`\n\nPresent the summary:\n\n```\nRETROSPECTIVE — {PROJECT_NAME}\nPeriod: {start_date} → {today}\nContext: {Collaborative|Autonomous|Mixed} ({detail})\n\nDuring this period:\n  Specs completed: {N}\n  Commits: {N}\n  Agent learnings recorded: {N} ({N} project, {N} workstation)\n  Human directives captured: {N}\n  Away-mode sessions: {N}\n```\n\n### Phase 3: Human Directives (Table 1)\n\nExtract directives the human gave during the retro window from:\n- **Handoff files** — decisions, explicit instructions\n- **Learnings entries** with `classified_by: \"human\"` — human-reclassified knowledge\n- **`.add\u002Fobservations.md`** — entries recording human feedback or directives\n- **Conversation context** — explicit instructions from the current or recent sessions\n\nFor each directive, **classify scope:**\n\n| Scope | Signal |\n|-------|--------|\n| project | References specific files, config, routes, schemas unique to this project |\n| workstation | References tools, libraries, workflows that apply across projects |\n| organization | References team, org, or company patterns (stub — accept but note \"future tier\") |\n| community | References universal methodology insights (stub — accept but note \"future tier\") |\n\n**If no directives found:** Skip Table 1 entirely, move to Phase 4. Do not show an empty table.\n\n**If directives found:** Present:\n\n```\nIn our recent sessions, you provided some key insights:\n\n━━━ TABLE 1: YOUR DIRECTIVES ━━━\n| # | Directive | Scope | Source |\n|---|-----------|-------|--------|\n| 1 | {directive text} | {scope} | {source} |\n| 2 | {directive text} | {scope} | {source} |\n\nAre these captured correctly in learnings? Any to add, remove, or reclassify?\n```\n\nWait for human confirmation. Apply any changes to learnings JSON files.\n\n### Phase 4: Agent Observations (Tables 2 & 3)\n\nRead agent-generated entries since last retro:\n- **Table 2 (Project):** Entries from `.add\u002Flearnings.json` where `classified_by: \"agent\"` and `date` within retro window\n- **Table 3 (Workstation):** Entries from `~\u002F.claude\u002Fadd\u002Flibrary.json` where `classified_by: \"agent\"` and `date` within retro window\n\n**Skip any table that has zero entries.** Do not show empty tables.\n\n**ADD Methodology Adherence Self-Assessment:**\n\nInclude in the agent observations a self-assessment of how well ADD methodology was followed during the retro window. Check each rule:\n\n| Rule | How to Assess |\n|------|---------------|\n| Spec-before-code | Were any features implemented without a spec? Check git log for implementation commits vs spec dates |\n| TDD cycles | Were tests written before implementation? (N\u002FA for markdown-only plugins) |\n| Auto-handoffs | Were handoffs written after commits and major work? Check `.add\u002Fhandoff.md` timestamps vs git log |\n| Learning checkpoints | Were learnings recorded at trigger points (post-verify, post-tdd, post-deploy, post-away)? Count expected vs actual |\n| Quality gates | Were `\u002Fadd:verify` runs done? Did they pass first time? |\n| Source control | Were conventional commits used? Check last N commits for pattern compliance |\n\nFormat methodology adherence as a **table** (not inline checkmarks — tables are easier to scan):\n\n```\n━━━ TABLE 2: AGENT OBSERVATIONS (PROJECT) ━━━\n| # | Observation | Severity |\n|---|-------------|----------|\n| 1 | {observation} | {severity} |\n\nADD Methodology Adherence:\n| Rule | Status | Detail |\n|------|--------|--------|\n| Spec-before-code | ✓ Pass | {assessment} |\n| Auto-handoffs | ✓ Pass | {assessment} |\n| Learning checkpoints | ✗ Fail | {assessment with specifics} |\n```\n\n**Do NOT show Table 3 if it has zero entries.** Skip it silently — no heading, no empty table, no mention.\n\nThen ask:\n\n```\nHelp me polish these. Do you disagree or wish to modify any of these learnings?\n```\n\nWait for human input. If the human identifies data gaps (missing checkpoints, incorrect entries), fix them **in the background** — queue the writes and continue the retro flow immediately. Do not block the human while writing JSON entries or regenerating markdown views.\n\n### Phase 5: Targeted Questions\n\nAsk one at a time. Adapt based on session context:\n\n**Q1 (skip if Autonomous context):** \"What went well?\"\n→ Only ask if the human had enough interactive context to answer meaningfully.\n\n**Q2:** \"What needed improvement that was not included already in our learnings?\"\n→ Scoped to gaps — the tables already surfaced known issues.\n\n**Q3 (rate-limited — 1x per calendar day):** \"On a scale of 0.0 to 9.0, how well are we working together?\"\n→ To check rate limit: read `.add\u002Fretros\u002F` for any file matching `retro-{today's date}*.md` that already has a `Human Collaboration` score populated. If found, skip this question.\n\n**Q4 (rate-limited — 1x per calendar day):** \"Any improvements for Agent Driven Development you would suggest?\"\n→ Same rate-limit check as Q3. These two are always asked\u002Fskipped together.\n\n### Phase 6: Agent Self-Assessment Scores\n\nThe agent provides two self-assessed scores with evidence. These are NOT asked of the human — the agent generates them from data.\n\n**ADD Methodology Effectiveness (0.0-9.0):**\n- Based on methodology adherence from Phase 4\n- Consider: spec coverage, TDD compliance, handoff discipline, checkpoint completeness, quality gate usage\n- Must cite specific evidence (not vague claims)\n\n**Swarm Effectiveness (0.0-9.0):**\n- How well did agents collectively build together?\n- Consider: parallel subagent usage, context handoff quality between sessions, duplicate work avoided, session continuity\n- Must cite specific evidence\n\nFormat:\n\n```\n━━━ AGENT SELF-ASSESSMENT ━━━\nADD Methodology Effectiveness: {X.X} \u002F 9.0\n  Evidence: {specific evidence from the retro window}\n\nSwarm Effectiveness: {X.X} \u002F 9.0\n  Evidence: {specific evidence from the retro window}\n```\n\nScores must be justified by evidence. If the agent detects its own score seems inflated relative to evidence, adjust downward. Honesty over optimism.\n\n### Phase 7: Record and Update\n\n1. **Write retro archive:**\n   Use `${CLAUDE_PLUGIN_ROOT}\u002Ftemplates\u002Fretro.md.template` as structure.\n   Write to `.add\u002Fretros\u002Fretro-{date}.md` (create directory if needed).\n   Fill all sections with data from the retro.\n\n2. **Store scores:**\n   Read `.add\u002Fretro-scores.json` (create from `${CLAUDE_PLUGIN_ROOT}\u002Ftemplates\u002Fretro-scores.json.template` if doesn't exist).\n   Append entry:\n   ```json\n   {\n     \"date\": \"{YYYY-MM-DD}\",\n     \"collab_score\": {X.X},\n     \"add_effectiveness\": {X.X},\n     \"swarm_effectiveness\": {X.X},\n     \"retro_file\": \".add\u002Fretros\u002Fretro-{date}.md\",\n     \"context\": \"{collaborative|autonomous|mixed}\"\n   }\n   ```\n   If collab score was rate-limited (skipped), use `null` for `collab_score`.\n\n3. **Store ADD feedback:**\n   If the human provided ADD improvement suggestions, append to `.add\u002Fadd-feedback.md`:\n   ```markdown\n   ## {YYYY-MM-DD}\n   - **Suggestion:** {text}\n   - **Retro:** .add\u002Fretros\u002Fretro-{date}.md\n   - **Streamed:** false\n   ```\n   Create the file with `# ADD Methodology Feedback` header if it doesn't exist.\n\n4. **Update learnings:**\n   Write new learning entries from the retro as structured JSON to the appropriate file (`.add\u002Flearnings.json` or `~\u002F.claude\u002Fadd\u002Flibrary.json`). Follow the checkpoint process in `${CLAUDE_PLUGIN_ROOT}\u002Freferences\u002Flearning-reference.md`:\n   - Classify scope for each new learning\n   - Write to the appropriate JSON file\n   - Regenerate the corresponding markdown view\n\n5. **Scope review and reclassification:**\n   Review entries classified by agents since the last retro:\n   ```\n   SCOPE REVIEW — Agent-classified entries since last retro:\n     L-{NNN}: \"{title}\" — classified as {scope} by agent\n     WL-{NNN}: \"{title}\" — classified as {scope} by agent\n\n   Any reclassifications needed?\n   ```\n   For each reclassification the human approves:\n   - Move the entry between JSON files\n   - Update `scope` field and `classified_by` to `\"human\"`\n   - Assign new ID appropriate to target file\n   - Regenerate both markdown views\n\n   **Auto-promotion proposals (batch):**\n\n   In addition to asking per-entry, scan `.add\u002Flearnings.json` entries since the last retro and auto-propose **workstation promotion candidates** — entries that look universal enough to belong in `~\u002F.claude\u002Fadd\u002Flibrary.json` for cross-project use.\n\n   Promotion signals (ANY of):\n   - `severity >= \"high\"` AND `category = \"anti-pattern\"`\n   - Body contains \"always\", \"never\", \"every time\", \"on every\", \"default\" — language indicating a general rule rather than a project-specific fact\n   - `stack` entries describe tooling that the user's other projects in `~\u002F.claude\u002Fadd\u002Fprojects\u002F` also use (check registry stacks)\n   - Body references generic cloud\u002Finfra primitives (GCE, AWS, GKE, EKS, ArgoCD, cert-manager, etc.) without project-specific identifiers\n\n   Present the candidates as one batch table:\n\n   ```\n   PROMOTION CANDIDATES — cross-project workstation learnings\n\n   | # | Entry | Signal | Promote? |\n   |---|-------|--------|----------|\n   | 1 | L-{NNN}: \"{title}\" | {signal that matched} | [y\u002Fn] |\n   | 2 | L-{NNN}: \"{title}\" | {signal that matched} | [y\u002Fn] |\n   ...\n\n   Reply with the numbers to promote (e.g., \"1,3,5\" or \"all\" or \"none\").\n   ```\n\n   For each approved promotion:\n   - Copy entry to `~\u002F.claude\u002Fadd\u002Flibrary.json` with a new `WL-{NNN}` ID\n   - Set `source` to the project name\n   - Set `classified_by` to `\"human\"`\n   - Leave original entry in `.add\u002Flearnings.json` (projects may still reference it)\n   - Regenerate both markdown views at the end\n\n   Why this exists: the agentVoice retro (2026-04-12) surfaced 8 clearly-universal lessons (ArgoCD, CI-disabled, Cloud Build, WebSocket timeout, E2E skip ban) that sat in project markdown for 40 days without promotion. A batch proposal at retro time closes that gap.\n\n6. **Update cross-project persistence:**\n   a. **Profile updates** (`~\u002F.claude\u002Fadd\u002Fprofile.md`): If retro reveals preferences that carry to other projects, ask: \"Add to your ADD profile?\"\n   b. **Project index** (`~\u002F.claude\u002Fadd\u002Fprojects\u002F{name}.json`): Update `last_retro` date and `learnings_count`.\n\n7. **Promote to plugin-global (ADD dev project only):**\n   If running inside the ADD plugin project (detected by `knowledge\u002Fglobal.md` existing as a local file), present Tier 1 promotion candidates.\n\n8. **Apply config\u002Ftemplate changes:** If agreed changes affect the process, make edits now.\n\n9. **Deduplicate knowledge stores:**\n   Check all stores for duplicates or misplaced entries. Report: \"{N} duplicates consolidated, {N} entries relocated.\"\n\n10. **Prune stale entries:**\n    - Observations >30 days old without `[synthesized M-{NNN}]` → archive\n    - Learnings >90 days old without references → flag for human review (never auto-delete)\n\n11. **Regenerate markdown views** after all JSON modifications.\n\n### Phase 8: Observation Synthesis\n\nRead `.add\u002Fobservations.md` and synthesize:\n\n1. Group observations by operation type\n2. Identify patterns (3+ similar = a pattern)\n3. For each pattern, propose a process mutation:\n   ```\n   ### Proposed Mutation: {title}\n   **Skill:** \u002Fadd:{skill-name}\n   **Change:** {concrete change to the skill}\n   **Evidence:** {observation timestamps and summaries}\n   **Expected outcome:** {what should improve}\n   ```\n4. Present proposals to human for approval. Only apply approved mutations.\n\n### Phase 9: Apply Approved Mutations\n\nFor each human-approved mutation:\n1. Read the target skill's SKILL.md\n2. Apply the change\n3. Log in `.add\u002Fmutations.md`\n4. Mark source observations as synthesized\n\n### Phase 10: Process Health Assessment\n\nReview `.add\u002Fmutations.md` for previously applied mutations:\n1. Did the problem recur? → Strengthen mutation\n2. Did the problem stop? → Note positive outcome\n3. New side effects? → Adjust mutation\n4. Report process health summary\n\n### Phase 11: Maturity Promotion Assessment\n\nIf the retro surfaces a promotion request, run an evidence-based check. See maturity promotion criteria:\n\n**POC → Alpha:** At least 3 evidence items (specs, tests, CI, commits, etc.) + core concept validated\n**Alpha → Beta:** Feature specs exist, coverage >50%, CI\u002FCD configured, PR workflow, 2+ environments, conventional commits, TDD evidence\n**Beta → GA:** Coverage >80%, protected branches, release tags, 3+ environments, all gates blocking, 30+ days stability, SLAs defined\n\nDo NOT promote without evidence. Promotion requires proof, not aspiration.\n\n---\n\n## Mode 2: Agent Summary (`--agent-summary`)\n\nQuick, non-interactive mode. Present accumulated observations without a full retro.\n\n1. Read `.add\u002Flearnings.json` and `~\u002F.claude\u002Fadd\u002Flibrary.json`\n2. Filter entries since last retro\n3. Present structured summary:\n\n```\nAGENT OBSERVATIONS — since {last_retro_date}\n\nTECHNICAL DISCOVERIES ({N} entries):\n  - {discovery with date}\n\nARCHITECTURE DECISIONS ({N} entries):\n  - {decision with rationale}\n\nADD METHODOLOGY ADHERENCE:\n  {self-assessment checklist}\n\nWHAT WORKED:\n  - {positive pattern}\n\nWHAT DIDN'T:\n  - {negative pattern}\n\nSUGGESTED CHANGES:\n  1. {suggestion with rationale}\n\nThese are observations only — no changes applied.\nRun \u002Fadd:retro for a full interactive retrospective.\n```\n\n---\n\n## Retro Frequency Guidance\n\nDisplay during the first retro:\n\n```\nRECOMMENDED RETRO CADENCE:\n  - Agent auto-checkpoints: Continuous (after each verify, cycle, away session)\n  - Agent summary (\u002Fadd:retro --agent-summary): Weekly or before planning\n  - Full interactive retro (\u002Fadd:retro): After each feature\u002Fsprint completion,\n    or every 2 weeks — whichever comes first\n```\n\n---\n\n## Score Semantics\n\nAll scores use 0.0-9.0 scale with 1 decimal precision:\n\n| Range | Meaning |\n|-------|---------|\n| 0.0-2.0 | Poor — process not working, significant friction |\n| 2.1-4.0 | Below average — notable gaps, frequent workarounds |\n| 4.1-6.0 | Adequate — functional but room for improvement |\n| 6.1-8.0 | Good — effective with minor issues |\n| 8.1-9.0 | Excellent — highly effective, minimal friction |\n\n---\n\nEnd-of-skill epilogue: follow `${CLAUDE_PLUGIN_ROOT}\u002Freferences\u002Fskill-epilogue.md` (observation + learning checkpoint + progress tracking). The retro's Phase 7 already records its learnings — do not duplicate them; the epilogue's observation line still applies.\n",{"references":46,"description":7,"allowed-tools":51,"argument-hint":59},[47,48,49,50],"learning-reference.md","maturity-matrix.md","skill-epilogue.md","rules\u002Ftelemetry.md",[52,53,54,55,56,35,57,58],"Read","Write","Edit","Glob","Grep","AskUserQuestion","TodoWrite","[--agent-summary] [--since YYYY-MM-DD] [--scope feature|sprint|session] [--dry-run]","main","e335248c85d50f73919320bc138766eb413c5ee3","core\u002Fskills\u002Fretro",6,"https:\u002F\u002Fgithub.com\u002FMountainUnicorn\u002Fadd","https:\u002F\u002Fgetadd.dev","https:\u002F\u002Fgithub.com\u002FMountainUnicorn","2026-10-31 22:03:50",{"source":8,"skillName":5,"commands":69,"prompt":79,"zipAvailable":80},[70,73,76],{"tool":71,"command":72},"npx","npx skills add MountainUnicorn\u002Fadd --skill retro",{"tool":74,"command":75},"bunx","bunx skills add MountainUnicorn\u002Fadd --skill retro",{"tool":77,"command":78},"pnpm","pnpm dlx skills add MountainUnicorn\u002Fadd --skill retro","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FMountainUnicorn\u002Fadd\u002Fblob\u002Fmain\u002Fcore\u002Fskills\u002Fretro\u002FSKILL.md",true,[82,87,93],{"path":83,"resourceType":83,"resourceTypeDesc":84,"sizeBytes":85,"sourceHtmlUrl":86},"LICENSE","许可证",1071,"https:\u002F\u002Fgithub.com\u002Fjiayaoqijia\u002Fcryptoskill\u002Fblob\u002Fmain\u002Fskills\u002Fmcp-servers\u002Fdanthamanvoiss-video-creation-router\u002FLICENSE",{"path":88,"resourceType":89,"resourceTypeDesc":90,"sizeBytes":91,"sourceHtmlUrl":92},"SOURCE.md","REFERENCE","参考文档",418,"https:\u002F\u002Fgithub.com\u002Fjiayaoqijia\u002Fcryptoskill\u002Fblob\u002Fmain\u002Fskills\u002Fmcp-servers\u002Fdanthamanvoiss-video-creation-router\u002FSOURCE.md",{"path":94,"resourceType":89,"resourceTypeDesc":90,"sizeBytes":95,"sourceHtmlUrl":96},"TRUST.auto.yaml",2217,"https:\u002F\u002Fgithub.com\u002Fjiayaoqijia\u002Fcryptoskill\u002Fblob\u002Fmain\u002Fskills\u002Fmcp-servers\u002Fdanthamanvoiss-video-creation-router\u002FTRUST.auto.yaml",[98,106,111,116,126,152],{"skillNo":99,"name":100,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"publishedAt":41},"01M6CHXN68ZCGNXE9Q0K2T2RKX","add-verify","https:\u002F\u002Fgithub.com\u002FMountainUnicorn\u002Fadd\u002Fblob\u002Fmain\u002Ftests\u002Fhooks\u002Ffixtures\u002Fdoctor\u002Fmanifest-modified\u002Fskills\u002Fadd-verify\u002FSKILL.md",[],[104,105],"add verify","verify",{"skillNo":107,"name":100,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":108,"categories":109,"keywords":110,"publishedAt":41},"01M6CHXJY4ZXKTNP0235A9KTNH","https:\u002F\u002Fgithub.com\u002FMountainUnicorn\u002Fadd\u002Fblob\u002Fmain\u002Ftests\u002Fhooks\u002Ffixtures\u002Fdoctor\u002Fmanifest-missing\u002Fskills\u002Fadd-verify\u002FSKILL.md",[],[104,105],{"skillNo":112,"name":100,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":113,"categories":114,"keywords":115,"publishedAt":41},"01M6CHXG6NV1DBMWT2N6XE5555","https:\u002F\u002Fgithub.com\u002FMountainUnicorn\u002Fadd\u002Fblob\u002Fmain\u002Ftests\u002Fhooks\u002Ffixtures\u002Fdoctor\u002Fhealthy\u002Fskills\u002Fadd-verify\u002FSKILL.md",[],[104,105],{"skillNo":117,"name":26,"nameZh":6,"description":118,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":119,"categories":120,"keywords":121,"publishedAt":41},"01M6CHXDZFPEFBZ79GHP48S1BB","[ADD v0.11.0] Show installed version, project version, and upgrade status","https:\u002F\u002Fgithub.com\u002FMountainUnicorn\u002Fadd\u002Fblob\u002Fmain\u002Fplugins\u002Fadd\u002Fskills\u002Fversion\u002FSKILL.md",[],[26,32,122,40,35,23,123,31,124,125],"status","check","v0","show",{"skillNo":127,"name":105,"nameZh":6,"description":128,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":129,"categories":130,"keywords":140,"publishedAt":41},"01M6CHXB5XG415PP88E0GACHP8","[ADD v0.11.0] Run quality gates — lint, types, tests, coverage, spec compliance","https:\u002F\u002Fgithub.com\u002FMountainUnicorn\u002Fadd\u002Fblob\u002Fmain\u002Fplugins\u002Fadd\u002Fskills\u002Fverify\u002FSKILL.md",[131,134,135],{"categoryNo":132,"categoryName":133,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000011","运维与部署",{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},{"categoryNo":136,"categoryName":137,"taxonomyType":138,"taxonomyTypeDesc":139},"01JBSEEDCAT000000000000037","律师与法务","PROFESSION","适用职业",[105,32,31,141,142,124,143,144,23,40,145,146,38,147,148,149,150,35,34,151],"coverage","fix","smoke","gate","spec","quality gates","lint","ci","compliance","secrets","quality",{"skillNo":153,"name":154,"nameZh":6,"description":155,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":156,"categories":157,"keywords":165,"publishedAt":41},"01M6CHX8JF3M3J8BKTBW604D37","ux","[ADD v0.11.0] Iterate on UI\u002FUX design before implementation — wireframes, flow validation, and design sign-off","https:\u002F\u002Fgithub.com\u002FMountainUnicorn\u002Fadd\u002Fblob\u002Fmain\u002Fplugins\u002Fadd\u002Fskills\u002Fux\u002FSKILL.md",[158,161,162],{"categoryNo":159,"categoryName":160,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000015","视觉设计",{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},{"categoryNo":163,"categoryName":164,"taxonomyType":138,"taxonomyTypeDesc":139},"01JBSEEDCAT000000000000032","设计师",[154,166,167,168,169,40,145,35,170,23,32,31,30,29,124,171,172,173,174],"ui","file","figma","validation","system","design-system","implementation","flow","design"]