think

name: think description: "Use when entering deep thinking mode, exploring ideas, finding patterns, or reflecting on vault content. Also triggers on: challenge, stress-test, emerge, drift, trace, connect, ideas, ghost, brainstorm, what if, poke holes" auto_improve: false # divergent/introspective skill — never auto-patch from corrections (thinking/aha/2026-04-12-thin-skills-must-not-auto-improve.md). Evolve by manual decision only.

Pre-Think Check

IMPORTANT: Before starting, check for un-audited knowledge. Run: python3 ${CLAUDE_PLUGIN_ROOT}/skills/audit/scripts/audit.py {vault}/knowledge/raw --status If any book shows ❌ or 👁️, warn: "⚠️ Un-audited knowledge: [book]. Run /audit to verify."

Vault Location

Vault path: Read from ${CLAUDE_PLUGIN_ROOT}/brain-os.config.md

/think — Deep Thinking Mode

Usage

/think                        Open-ended reflection and exploration
/think digest                 Review agent-output, promote insights to proper vault zones
/think challenge <topic>      Stress-test a belief against vault evidence
/think emerge                 Surface ideas the vault implies but never stated
/think drift                  Compare stated goals vs actual behavior (30-60 days)
/think trace <topic>          Track how an idea evolved over time
/think connect <A> <B>        Find bridges between two domains
/think ideas                  Brainstorm across all vault domains
/think ghost <question>       Answer a question as the user would

If no angle given, show the menu above and ask: "What's on your mind?"

General Behaviors (all angles)

• Focus on {vault}/thinking/, {vault}/daily/, and {vault}/context/
• Minimize business framing unless user brings it up
• Be a thinking partner — ask probing questions, challenge assumptions
• Write output to {vault}/thinking/agent-output/YYYY-MM-DD-<angle>-<topic>.md
• Ask "why" and "what if" questions, surface contradictions, suggest new angles

Angles

digest

Review agent-output files and promote insights to proper thinking/ zones:

1. Scan — read all files in {vault}/thinking/agent-output/ (skip README.md)
2. Summarize each file — 2-3 line summary per file, grouped by theme
3. Present for review — show the user what's there:

   📂 Agent Output (N files)
   
   1. [filename] — summary
      → Suggested zone: patterns/ | ideas/ | connections/ | reflections/ | principles/
      → Key insight: one-liner
   
   2. [filename] — summary
      → Suggested zone: ...
      → Key insight: ...
   

4. Ask user — "Which insights to promote? (all / numbers / skip)"
5. For each promoted insight:
   • Move file from agent-output/ to the target zone (rename if needed for clarity)
   • Update frontmatter: add promoted: YYYY-MM-DD, update zone field
   • Add wiki-links from related existing vault files → the promoted file
   • Add wiki-links from the promoted file → related vault files
6. Clear agent-output — after all files are processed (promoted, merged, or skipped), delete ALL processed files from agent-output/. This folder is a staging area, not storage. Knowledge already lives in its destination.
7. Summary — report what was promoted where, with link count

Rules:

• NEVER auto-promote without user confirmation — this is the human's thinking zone
• Suggest zones based on content type: recurring theme → patterns/, novel idea → ideas/, bridge between domains → connections/, personal observation → reflections/, reusable principle → principles/
• If an insight overlaps with an existing vault file, suggest merging instead of creating a new file — then delete the agent-output file
• Keep promoted filenames clean: YYYY-MM-DD-topic.md (drop the grill-insights- prefix)
• agent-output/ must be empty (except README.md) after digest completes — no exceptions

challenge

Pressure-test a belief or decision:

1. Find current position — scan vault for what user believes about this topic
2. Find contradictions — stated beliefs that conflict with actions or other beliefs
3. Present counter-evidence — arguments, data, perspectives that challenge the view
4. Track belief shifts — has user changed mind on this before?
5. Ask hard questions — 3-5 questions probing the weakest parts

emerge

Surface ideas the vault implies but user never explicitly stated:

1. Deep scan — read across thinking/, daily/, personal/research/, context/
2. Find scattered premises — ideas in different contexts that form a larger conclusion
3. Identify unnamed patterns — recurring themes not yet named or formalized
4. Surface unarticulated directions — where actions/interests point that user hasn't consciously decided

drift

Compare stated intentions vs actual behavior:

1. Read goals — context/goals.md, context/strategy.md
2. Read 30-60 days of daily notes — what was actually spent time on
3. Compare — where do goals align with activity? Where do they diverge?
4. Surface avoidance — goals set but consistently not worked on
5. Surface hidden priorities — things done repeatedly but not in stated goals

trace

Track how an idea evolved over time:

1. Build vocabulary map — search vault for all mentions and related terms
2. Create timeline — order mentions chronologically
3. Identify phases — discovery, skepticism, adoption, mastery, etc.
4. Show evolution — how understanding or position changed
5. Current state — where is the user now with this idea?

connect

Find bridges between two domains:

1. Scan domain 1 — find all vault files related
2. Scan domain 2 — find all vault files related
3. Find bridges — shared themes, patterns, insights
4. Generate novel connections — non-obvious links based on vault evidence
5. Suggest actions — what could be done with these connections?

ideas

Comprehensive idea generation:

1. Gather vault structure — files, connections, orphans, dead ends
2. Read context — context/, recent daily/, active projects/
3. Scan thinking zone — ideas/, patterns/, connections/
4. Cross-domain analysis — patterns across different vault areas
5. Generate ideas: tools to build, content to create, systems to implement, conversations to have, subjects to investigate
6. Rank by impact — top 5 "do now" items

ghost

Answer a question as the user would:

1. Build voice profile — read about-me.md, preferences.md, recent daily notes, reflections/
2. Scan relevant vault files for actual opinions on the topic
3. Write answer in user's voice — matching tone, vocabulary, reasoning, values
4. Rate fidelity — confidence 1-10 that this matches how user would respond
5. Flag gaps — where vault lacks data on this topic

Outcome log

Follow skill-spec.md § 11. Append to {vault}/daily/skill-outcomes/think.log:

{date} | think | {action} | ~/work/brain-os-plugin | thinking/agent-output/{date}-{angle}-{topic}.md | commit:{hash} | {result}

• action: think, digest, challenge, emerge, drift, trace, connect, ideas, or ghost
• result: pass if output written, partial if gaps flagged or vault data sparse, fail if no relevant vault content found
• Optional: args="{angle} {topic}"