agentleFS
AbilityaiAbilityai/cornelius/.claude/skills/manage-thinking-topics/SKILL.md

manage-thinking-topics

Skill109 starsChanged 39 days agoMITView on GitHub

Manage the incubation loop topic lifecycle — seed new questions, review status, crystallize converged conclusions, and retire stale topics

The file

Show its rules and text

We split this skill into the rules it gives an agent

22 rules. About 0 are common, 16 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈16

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run count at convergence

    line 58
  2. Very specific:

    If there are converged topics, highlight: "[N] topics ready for crystallization — run /manage-thinking-topics crystallize [slug] to …

    line 86
  3. Distinctive:

    Purpose: Add a new question to the incubation loop.

    line 92
  4. Distinctive:

    If the topic slug is not provided, ask for it.

    line 98
  5. Very specific:

    Check that Brain/05-Meta/Thinking/[topic-slug].md does NOT already exist.

    line 108
  6. Distinctive:

    Check that the slug is not already in the registry.

    line 110
  7. Distinctive:

    Add a row to the Active Topics table:

    line 161
  8. Very specific:

    Update updated and updated_by in frontmatter.

    line 167
  9. Very specific:

    Verify status: converged. If not converged, warn and stop.

    line 191
  10. Distinctive:

    Read the full file only if the last 200 lines do not contain a "Current Best Answer" or equivalent conclusion.

    line 208
  11. Distinctive:

    Check if a permanent note already exists covering this conclusion.

    line 216
  12. Distinctive:

    Before drafting, decide the destination based on the nature of the conclusion:

    line 220
  13. Very specific:

    Update updated and updated_by

    line 280
  14. Distinctive:

    Remove the row from Converged table

    line 294
  15. Distinctive:

    Add a row to Crystallized table:

    line 295
  16. Distinctive:

    If the existing Crystallized table column header is still "Permanent Note", update it to "Destination" so it accommodates both atomic notes …

    line 301
  17. Distinctive:

    Purpose: Remove a topic from active analysis without crystallizing it (topic no longer relevant, question was wrong, superseded by new …

    line 316
  18. Distinctive:

    Remove the row from whichever table it was in (Active or Converged)

    line 354
  19. Distinctive:

    Do NOT add to Crystallized table (it wasn't graduated)

    line 355
  20. Very specific:

    When seeding or reviewing, the next analytical move is determined by run_count mod 6:

    line 379
  21. Distinctive:

    No major open questions flagged in last run

    line 398
  22. Distinctive:

    After completing this skill's primary task, consider tactical improvements:

    line 418

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Manage Thinking Topics

Human-in-the-loop manager for the incubation loop topic registry. Handles the full lifecycle: seeding new questions into analysis, reviewing current status, graduating converged conclusions to permanent notes, and retiring topics no longer worth pursuing.

Companion skill: .claude/skills/incubation-loop/SKILL.md — the analytical engine that processes active topics.

State Dependencies

SourceLocationReadWriteDescription
Thinking RegistryBrain/05-Meta/Thinking/THINKING-REGISTRY.md✓✓Master list of all topics + status
Topic FilesBrain/05-Meta/Thinking/[topic-slug].md✓✓Per-topic reasoning journal
Permanent NotesBrain/02-Permanent/✓Atomic crystallized conclusions (graph-integrated, future-reasoning substrate)
Synthesis ReportsBrain/05-Meta/Reports/✓Terminal crystallized conclusions (decision-input snapshots, excluded from LBS index)
AI Extracted NotesBrain/AI Extracted Notes/✓Alternative destination for crystallized notes

Modes

Invoke with one of four modes. If no mode is provided, default to review.


Mode: review

Purpose: Show the full topic landscape at a glance - what's active, what needs crystallization, what's done.

Step 1: Read Registry

date '+%Y-%m-%d'

Read Brain/05-Meta/Thinking/THINKING-REGISTRY.md.

Step 2: For Each Active Topic - Load Summary

For each topic with status: active, read its thinking file. Extract:

  • run_count - how many iterations completed
  • last_run - date of last iteration
  • Current leading hypothesis + confidence score
  • Next move in rotation (run_count mod 6)
  • Any open questions flagged

Step 3: For Each Converged Topic - Load Summary

For each topic with status: converged, read its thinking file. Extract:

  • The converged conclusion (Current Best Answer from last run)
  • Confidence score at convergence
  • Run count at convergence

Step 4: Present Status Report

Output a structured report:

## Thinking Topics: Status Review [DATE]

### Active Topics (N)
| Topic | Runs | Last Run | Leading Hypothesis | Confidence | Next Move |
|-------|------|----------|-------------------|------------|-----------|

### Converged — Ready for Crystallization (N)
| Topic | Runs | Conclusion Summary | Confidence |
|-------|------|-------------------|------------|

### Crystallized — Done (N)
| Topic | Permanent Note | Crystallized |
|-------|---------------|-------------|

### Summary
- [N] topics in active analysis
- [N] topics ready for crystallization (action needed)
- [N] topics fully crystallized
- Next incubation-loop run: processes all active topics

If there are converged topics, highlight: "[N] topics ready for crystallization — run /manage-thinking-topics crystallize [slug] to graduate each one."


Mode: seed

Purpose: Add a new question to the incubation loop.

Argument: topic slug (e.g., agent-memory-architecture)

Step 1: Gather Requirements

If the topic slug is not provided, ask for it. Then gather:

  1. Central question - The exact question being analyzed. Precise framing matters - vague questions stay vague.
  2. Why it matters - 1-2 sentences on stakes or relevance.
  3. Initial hypotheses - At least 2 competing hypotheses with rough initial confidence percentages (must sum to ~100%).
  4. Known evidence - Any notes or sources already known to be relevant (optional).
  5. Constraints/assumptions - What's taken as given? What's out of scope? (optional)

Step 2: Validate Slug

Check that Brain/05-Meta/Thinking/[topic-slug].md does NOT already exist. If it does, warn and stop.

Check that the slug is not already in the registry. If it is, warn and stop.

Step 3: Create Topic File

date '+%Y-%m-%d'

Create Brain/05-Meta/Thinking/[topic-slug].md:

---
created: YYYY-MM-DD
updated: YYYY-MM-DD
created_by: claude-sonnet-4-6
updated_by: claude-sonnet-4-6
agent_version: 01.25
topic: "[exact question being analyzed]"
run_count: 0
last_run: null
status: active
---

# Thinking: [Topic Title]

## Central Question
[The exact question. Precise framing matters - ambiguous questions stay ambiguous.]

## Why This Question Matters
[1-2 sentences on stakes or relevance]

## Initial Hypotheses
| Hypothesis | Initial Confidence | Basis |
|-----------|-------------------|-------|
| H1: [statement] | X% | [prior knowledge or intuition] |
| H2: [statement] | X% | [prior knowledge or intuition] |

## Known Evidence (Pre-Run)
[Any notes or sources already known to be relevant, or "None identified yet."]

## Constraints and Assumptions
[What are you taking as given? What's out of scope?]

---
*Analytical runs will be appended below by the incubation loop*

Step 4: Add to Registry

Read Brain/05-Meta/Thinking/THINKING-REGISTRY.md.

Add a row to the Active Topics table:

| [topic-slug] | [central question] | active | 0 | null |

Update updated and updated_by in frontmatter.

Step 5: Confirm

Report:

  • Topic file created at Brain/05-Meta/Thinking/[topic-slug].md
  • Registry updated - topic is now active
  • First incubation-loop run will apply: ACH Audit (move 0)
  • Estimated convergence: 4-6 runs (~4-6 days at daily schedule)

Mode: crystallize

Purpose: Graduate a converged topic — either as an atomic permanent note (Brain/02-Permanent/, graph-integrated) or as a terminal synthesis report (Brain/05-Meta/Reports/, excluded from semantic indexing). Choose based on whether the conclusion is generative (future reasoning will reuse it) or terminal (snapshot of what was concluded).

Argument: topic slug of a converged topic

If no slug is provided: read the registry, present the converged topics (with one-line conclusions), and ask which to crystallize before proceeding. If prior session context points at a specific topic or pair (e.g., a just-completed analysis that named them), recommend it as the first option. Accept common misspellings of the mode name (e.g., "crystalize").

Step 1: Load Converged Topic

Read Brain/05-Meta/Thinking/[topic-slug].md.

Verify status: converged. If not converged, warn and stop.

Extraction strategy (prefer the canonical block, fall back gracefully):

  1. First, look for ## Final Synthesis - a canonical block appended when the loop transitions a topic to converged. If present, this is the authoritative source. Extract from it:
    • Conclusion (the substantive answer)
    • Confidence and leading hypothesis
    • Key evidence (already curated)
    • Remaining uncertainty
    • Related topics (if any)
  2. If no Final Synthesis block exists (legacy topics from before the standardized format), read only the last 200 lines of the file rather than the full thinking journal (some are 5,000+ lines). Extract:
    • The central question (from frontmatter)
    • The most recent "Current Best Answer" section
    • The most recent "Updated Hypotheses" table for confidence scores
    • Any "Open Questions" flagged in the final run
  3. Read the full file only if the last 200 lines do not contain a "Current Best Answer" or equivalent conclusion.

Step 2: Search for Existing Notes

python3 resources/local-brain-search/search.py "[central question keywords]" --limit 5 --mode spreading 2>/dev/null

Check if a permanent note already exists covering this conclusion. If so, note the overlap and ask whether to update it or create a new note.

Step 3: Choose Destination

Before drafting, decide the destination based on the nature of the conclusion:

SignalDestination
Conclusion is one atomic insight that future reasoning will citeBrain/02-Permanent/ (permanent note)
Conclusion has multiple co-equal findings that belong together (e.g., a four-stage architecture, a stratified framework)Brain/05-Meta/Reports/ (synthesis report)
Conclusion is a snapshot of a moment (geopolitical assessment, project-stack analysis) — decision-input, not future-reasoning substrateBrain/05-Meta/Reports/ (synthesis report)
Conclusion captures a generative principle that will be reasoned FROM in other contextsBrain/02-Permanent/ (permanent note)
Topic is transitory — relevant for current decisions, unlikely to be revisitedBrain/05-Meta/Reports/ (synthesis report)

[APPROVAL GATE] Propose the destination and ask the user to confirm or override:

"Recommend: [destination] because [reason]. Confirm, or override to [other destination]?"

Synthesis reports do not participate in connection-finder or semantic search (the Reports/ folder is excluded from LBS indexing). They are visible in Obsidian and can link OUT to permanent notes via [[wiki-links]], but permanent notes should rarely link IN to reports.

Step 4: Draft Content

Adapt the draft to the chosen destination:

If permanent note (Brain/02-Permanent/):

  • Written in the user's voice, not as a report
  • Leads with the conclusion, not the process
  • Cites the key evidence that drove the conclusion
  • Notes remaining uncertainty or caveats
  • Atomic — one main insight, ~50-100 lines

If synthesis report (Brain/05-Meta/Reports/):

  • Long-form is appropriate — can be 200-600+ lines
  • Lead with executive summary, then layered detail
  • Multiple findings organized by section (each finding need not be atomic)
  • Cite full evidence base with sources/URLs where applicable
  • Wiki-link OUT to relevant permanent notes; reports are terminal

Propose the draft to the user for review before writing.

[APPROVAL GATE] Present the draft. Ask: "Write this to [chosen destination path]? (Yes / revise the draft / skip crystallization)"

Step 5: Write to Destination

On approval, write the file with proper frontmatter:

---
created: YYYY-MM-DD
updated: YYYY-MM-DD
created_by: claude-sonnet-4-6
updated_by: claude-sonnet-4-6
agent_version: 01.25
provenance: ai-inferred
---

Provenance (mandatory): Crystallized conclusions are agent-synthesized, so default provenance: ai-inferred. Per CLAUDE.md's guarded boundary, this only becomes endorsed or originated through an explicit endorsement act by the user at the [APPROVAL GATE] above - never set it automatically. If the user states the conclusion is his own thinking, set provenance: endorsed (or originated if it is purely his idea the loop merely formalized).

For synthesis reports, kebab-case the filename; add a date suffix if the report is a dated snapshot (e.g., trinity-production-architecture-2026-05-26.md).

Step 6: Update Topic File

Edit Brain/05-Meta/Thinking/[topic-slug].md:

  • Set status: crystallized
  • Update updated and updated_by
  • Append a final note (adapt to destination):
---

**CRYSTALLIZED [DATE]** → [[Destination Note or Report Name]]

*Conclusion graduated to [permanent knowledge base | synthesis reports]. Thinking file archived.*

Step 7: Update Registry

In Brain/05-Meta/Thinking/THINKING-REGISTRY.md:

  • Remove the row from Converged table
  • Add a row to Crystallized table:
| [topic-slug] | [[Destination Note or Report Name]] | YYYY-MM-DD |

If the existing Crystallized table column header is still "Permanent Note", update it to "Destination" so it accommodates both atomic notes and synthesis reports.

Update updated and updated_by in frontmatter.

Step 8: Confirm

Report:

  • File written at [full destination path]
  • Topic marked crystallized
  • Registry updated

Mode: retire

Purpose: Remove a topic from active analysis without crystallizing it (topic no longer relevant, question was wrong, superseded by new evidence).

Argument: topic slug

Step 1: Load Topic

Read Brain/05-Meta/Thinking/[topic-slug].md. Show:

  • Current status
  • Run count
  • Current leading hypothesis and confidence

Step 2: Confirm Retirement

[APPROVAL GATE] Ask:

"Retiring '[topic-slug]' will stop it from being processed by the incubation loop. The thinking file will be kept for reference but marked retired. Reason for retiring? (or type 'cancel')"

Capture the reason.

Step 3: Update Topic File

Edit Brain/05-Meta/Thinking/[topic-slug].md:

  • Set status: retired
  • Update updated and updated_by
  • Append:
---

**RETIRED [DATE]**

*Reason: [reason given]*

*Topic removed from active analysis. Thinking file kept for reference.*

Step 4: Update Registry

In Brain/05-Meta/Thinking/THINKING-REGISTRY.md:

  • Remove the row from whichever table it was in (Active or Converged)
  • Do NOT add to Crystallized table (it wasn't graduated)

Optionally add a Retired section if one doesn't exist:

## Retired

| Topic Slug | Reason | Retired On |
|------------|--------|------------|

Update updated and updated_by in frontmatter.

Step 5: Confirm

Report:

  • Topic marked retired in Brain/05-Meta/Thinking/[topic-slug].md
  • Removed from registry active/converged table
  • Thinking file preserved at full path for future reference

Move Rotation Reference

When seeding or reviewing, the next analytical move is determined by run_count mod 6:

run_count % 6Move
0ACH Audit — eliminate weak hypotheses
1Bayesian Update — explicit confidence tracking
2Steelman Opposition — dialectical challenge
3Cross-Domain Bridge — consilience from another field
4Implication Check — falsificationism
5Assumption Audit — find load-bearing shaky premises

Convergence Criteria (Reference)

A topic converges when ALL:

  • run_count >= 4
  • Same hypothesis has led for 3+ consecutive runs
  • Confidence delta between last two runs < 5 percentage points
  • No major open questions flagged in last run

The incubation-loop skill checks this automatically after each run.


Error Handling

ErrorRecovery
Registry missingStop; inform user to seed registry via incubation-loop skill
Topic file missing for registry entryLog warning; continue with other topics
Slug already exists (seed mode)Stop; show existing file path
Topic not converged (crystallize mode)Stop with current run count and status
KB search returns emptySkip search step; note in output

Self-Improvement

After completing this skill's primary task, consider tactical improvements:

  • Review execution: Were there friction points, unclear steps, or inefficiencies?
  • Identify improvements: Could step ordering, prompts to the user, or output formatting be clearer?
  • Scope check: Only execution changes — NOT changes to the incubation-loop analytical framework itself
  • Apply improvement (if identified):
    • Edit this SKILL.md with the specific improvement
    • Keep changes minimal and focused
  • Version control:
    • git add .claude/skills/manage-thinking-topics/SKILL.md
    • git commit -m "refactor(manage-thinking-topics): <brief improvement>"

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.