メニュー
Knowledge consolidation — transform session learnings into permanent architecture
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
SOC 職業分類に基づく
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/fabioc-aloha/alex-cognitive-architecture --skill meditationコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
Run the 7-step Artificial Critical Thinking pass — Materiality → Hypothesise → Alternatives → Disconfirmers → Audit priors → Severity → Commit-with-marker
Challenge what you think is right — alternative hypotheses, missing data, evidence quality, bias detection, falsifiability, and adversarial review
Step-back protocol — restate, generalise, specialise, invert, ask why, pre-mortem, check stakeholders, and audit framings before solving
| type | skill |
| lifecycle | stable |
| inheritance | inheritable |
| name | meditation |
| description | Knowledge consolidation — transform session learnings into permanent architecture |
| tier | standard |
| disable-model-invocation | true |
| applyTo | **/*meditat*,**/*reflect*,**/*consolidat* |
| currency | 2026-04-25T00:00:00.000Z |
Domain knowledge for guided knowledge consolidation sessions.
Meditation is the primary cognitive ritual. All other rituals are subordinate:
| Ritual | Relationship to Meditation |
|---|---|
| Dream | Diagnostic tool — chains after meditation sometimes |
| Self-Actualize | Deep meditation — monthly, includes Dream baseline first |
After completing the 5 R's, optionally trigger a Dream diagnostic when:
Dream produces the diagnostic report; meditation decides what to fix.
The inverse handoff also exists: a meditation may open by reading an existing dream-report.json and using it to focus the session. See dream-state/SKILL.md §Escalation for the finding→resolution mapping. This pattern replaces the former lucid-dream ritual (retired v8.4.0).
Review → Relate → Reinforce → Record → Resolve. The first four discover and persist; Resolve closes the meditation explicitly so the user knows what was saved and whether a dream follows.
/memories/ candidate| Signal |
|---|
| Keep? |
|---|
| Where |
|---|
| "I'll forget this by next week" | Yes | .instructions.md or SKILL.md |
| "This was non-obvious" | Yes | /memories/ or /memories/repo/ |
| "I keep rediscovering this" | Definitely | .instructions.md (procedural) |
| "This was specific to today" | No | Session memory only |
| "Everyone knows this" | No | Don't store |
| Type | Location |
|---|---|
| Repeatable process | .github/instructions/*.instructions.md |
| Domain expertise | .github/skills/*/SKILL.md |
| Workflow template | .github/prompts/*.prompt.md |
| Cross-project pattern | /memories/ (user memory) |
| Repo-specific fact | /memories/repo/ (repo memory) |
| Active multi-session task | /memories/session/ (conversation-scoped) |
| Long-term chronicle entry | .github/episodic/meditation-YYYY-MM-DD-{topic}.md |
.github/episodic/ is the project's long-term meditation/dream chronicle. Write an entry when:
dream-report-YYYY-MM-DD.md)File naming:
meditation-YYYY-MM-DD-{topic-slug}.md — meditation chronicledream-report-YYYY-MM-DD.md — dream diagnostic outputchronicle-YYYY-MM-DD-{name}.md — major milestone narrativeIf .github/episodic/ does not exist, create it. The folder is heir-portable.
Distinction from /memories/session/: episodic = permanent, project-scoped narrative; session = ephemeral, conversation-scoped state.
After recording insights, assess whether the welcome experience still fits:
| Question | If drifted | Config to update |
|---|---|---|
| Do taglines still reflect the project direction? | Regenerate | .github/config/taglines.json |
| Does the loop menu match actual workflows? | Add/remove groups | .github/config/loop-menu.json |
| Has the project phase changed? | Update phase | loop-menu.json → projectPhase |
| Is the identity/persona still accurate? | Revise identity layer | .github/copilot-instructions.md |
| Does the North Star still inspire? | Refresh vision | NORTH-STAR.md |
Trigger: Run this check when meditation surfaces project evolution, phase transitions, or domain pivots. Use the welcome-experience-customization skill for the full workflow.
Heir fallback: If taglines.json or loop-menu.json does not exist, skip this check — the project hasn't run welcome customization yet. Suggest it only if the user explicitly asks about welcome experience.
The 5th R closes the loop. Without it, meditations end ambiguously and the user is left wondering "did anything get saved?"
Required close steps:
Summarize persisted artifacts — list each file written/modified in this meditation, one line each:
Persisted:
- .github/instructions/foo.instructions.md (new — captures X pattern)
- /memories/patterns.md (added Y rule)
- .github/episodic/meditation-2026-04-25-bar.md (chronicle)
Evaluate dream chaining — explicitly check the criteria (see Dream Chaining section above):
.github/quality/dream-report.json does not exist, treat as never-dreamed (overdue). The file is created on first /dream run.State the verdict: "Chaining a dream now" / "Skipping dream — none of the criteria fired" / "Bookmarking dream for later".
One-line session summary — close with what changed: "Meditation persisted 3 insights across 2 surfaces; no dream needed."
After Resolve, the meditation is complete. Do not continue probing.
Don't just list — ask why it matters:
| Signal | Action |
|---|---|
| Repeating the same insight differently | Move on |
| No new connections after 3 attempts | Close topic |
| User energy dropping | Wrap up with quick record |
| Tangent detected | Bookmark for separate session |
meditation-snapshot.cjs)Path B (v8.4.0): chronicle files in .github/episodic/meditation-YYYY-MM-DD-*.md are the canonical record of when meditation last happened. The meditation-snapshot.cjs muscle scans them, derives the date from each filename (mtimes are unreliable across bulk git operations), and writes .github/quality/meditation-snapshot.json.
session-start.cjs reads this snapshot to decide whether to surface the "meditation overdue" nag — replacing the hand-edited cogConfig.lastMeditation field, which is now a fallback only.
| Field | Source |
|---|---|
lastMeditation | YYYY-MM-DD from newest matching filename |
lastMeditationFile | Workspace-relative path |
daysSinceLastMeditation | Computed from lastMeditation |
overdue | daysSinceLastMeditation >= 7 |
chronicleCount | Total matching chronicles |
cadence.last30Days / last90Days | Trailing-window counts |
pruneThreshold | 60 (D17) |
Run after authoring a new chronicle:
node .github/muscles/meditation-snapshot.cjs
The muscle prunes the episodic directory to the newest 60 chronicles (FIFO) on each non---json run. Older chronicles remain in git history; meditation's job is to produce new insight, not to curate the archive.
Four config files reference meditation as their lifecycle owner. The current honest state (v8.4.0):
| File | Claim | Status |
|---|---|---|
.github/config/assignment-log.json | "Meditation analyzes for routing patterns" | AFCP feature; pruning not yet automated. Manually inspect during meditation when the log has entries. |
.github/config/knowledge-artifacts.json | "Meditation prunes artifacts older than 90 days with confidence < 0.5" | Future trifecta — see PLAN-meditation-ritual-refactor.md. Manually curate during meditation for now. |
.github/config/unknowns.json | "Unknowns with state=resolved are pruned during meditation" | Manual curation during meditation. No automated pruning today. |
.github/config/session-metrics.json | "Session tracking for meditation analysis" | Gitignored local file; analyze in-session if useful. |
When meditating, scan these files only if they are non-empty.