/obsidian-learn

description: Review vault learnings, prune stale ones, surface active patterns - the vault's lessons compound or expire
category: thinking
triggers_en: ["review learnings", "what have I learned", "show lessons", "prune learnings"]
triggers_es: ["revisa los aprendizajes", "qué he aprendido", "muéstrame las lecciones", "poda los aprendizajes"]
triggers_pt: ["revise aprendizados", "o que eu aprendi", "mostre lições", "reduza aprendizados"]
triggers_zh: ["回顾我学到的东西", "我最近学到了什么", "整理知识库里的经验", "清理过时的学习记录"]

Use the obsidian-second-brain skill. Execute `/obsidian-learn $ARGUMENTS`:

The optional argument is a scope: `recent` (last 30 days, default), `all` (entire vault), or a topic name. Scope limits the review LIST only - the Stale and Superseded classifications always scan the entire vault, since a 6-months-old learning cannot appear inside a 30-day window.

1. Read `_CLAUDE.md` first if it exists in the vault root 2. Read `index.md` and recent operation log for vault context (if `Logs/` exists: read the last 2-3 `Logs/YYYY-MM-DD.md` files; otherwise read `log.md`)

3. Spawn parallel subagents to gather learnings:

• **Lessons agent**: scan all daily notes for "Lesson learned" sections, "What didn't" sections, evening review insights
• **Decisions agent**: read all ADRs in the decisions folder (resolved per `references/folder-map.md` - wiki-style `wiki/decisions/`, Obsidian-style `Knowledge/` with `ADR-` filename prefix) - extract the rationale and outcome of each
• **Reports agent**: read recent emerge/synthesize/connect/challenge reports in the concepts folder (resolved per `references/folder-map.md`) - the auto-generated pattern reports
• **Mistakes agent**: scan dev logs and daily notes for "what didn't work", "wasted time on", "next time", "lesson", phrases indicating learning from failure
• **Wins agent**: scan for patterns that worked - "this saved time", "this approach worked", recurring success patterns

4. For each learning found, classify:

• **Active**: still relevant, recurring, reinforced by recent activity
• **Stale**: 6+ months old with no recent reinforcement, or contradicted by newer evidence
• **Superseded**: explicitly replaced by a newer ADR or pattern
• **Promoted**: appeared 3+ times - should become a permanent rule in `_CLAUDE.md`

5. Generate the Learnings Report:

Active Learnings (still applies)

• List learnings reinforced in the last 90 days
• Cite the original source and most recent reinforcement

Stale Learnings (consider archiving)

• List learnings with no recent reinforcement
• Suggest: keep, archive, or convert to history note

Superseded Learnings (already replaced)

• Old position → New position with ADR reference

Promotion Candidates (appeared 3+ times)

• Learnings strong enough to become permanent rules in `_CLAUDE.md`
• Suggest exact wording for the operating manual

Top 5 Lessons of the Period

• Most impactful learnings ranked by frequency × recency × consequence

6. Save the report to the concepts folder (resolved per `references/folder-map.md`) as `YYYY-MM-DD - Learnings Review.md` (`type: learnings-review`, schema in `references/ai-first-rules.md`) 7. Append to the operation log: if `Logs/` exists write `**HH:MM** - learn | X active, Y stale, Z superseded, N promotion candidates` to `Logs/YYYY-MM-DD.md`; otherwise append `## [YYYY-MM-DD] learn | X active, Y stale, Z superseded, N promotion candidates` to `log.md` 8. Update today's daily note with a brief summary 9. Offer to:

• Promote candidates to `_CLAUDE.md` (with user confirmation)
• Archive stale learnings (with user confirmation)
• Export top 5 as a shareable markdown for content/journaling

Lessons that aren't reviewed don't compound. This command turns scattered notes into a living rulebook.

---

**AI-first rule:** Every note created or updated by this command MUST follow `references/ai-first-rules.md` - `## For future agent` preamble, rich frontmatter (`type`, `date`, `tags`, `ai-first: true`, plus type-specific fields), recency markers per external claim, mandatory `[[wikilinks]]` for every person/project/concept referenced, sources preserved verbatim with URLs inline, and confidence levels where applicable. If that path does not resolve from your working directory, search upward for it; if you still cannot read it, say so before writing rather than producing a note that silently skips the rule. The vault is for future agent retrieval - not human reading.

**Anti-fabrication:** Search exhaustively before claiming any note, person, or file is absent - false absence is the most common failure mode - and never invent facts, entities, or dates (mark unknowns as `TBD`). See the anti-fabrication and search-completeness hard rules in `references/ai-first-rules.md`.