agentleFS
aman-bhandariaman-bhandari/claude-code-agent-skills-framework/.claude/skills/brainstorm/SKILL.md

brainstorming-knowledge-builder

Skill0 starsChanged 7 months agoMITView on GitHub

Structured ideation and LLM knowledge base building using Karpathy's LLM Wiki architecture with Obsidian integration. Triggers on /brainstorm, BRAINSTORM: prefix, or when thinking through a problem, building knowledge, or ingesting sources. Three modes -- ideation (think through problems), discuss (create wiki from conversation), and knowledge (build persistent wiki from raw sources).

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    NEVER write code during brainstorming.

    line 30
  2. Very specific:

    Add [[wiki-links]] to every related concept page that exists.

    line 42
  3. Very specific:

    Update knowledge/wiki/index.md with the new/modified entry under the correct category.

    line 43
  4. Very specific:

    Add source to knowledge/raw/

    line 64
  5. Distinctive:

    NEVER write code during brainstorming

    line 73
  6. Distinctive:

    Ask the student to explain their reasoning at every decision point

    line 74
  7. Very specific:

    If brainstorming reveals a knowledge gap, flag it for /review-deck

    line 77
  8. Very specific:

    Use [[wiki links]] in knowledge files for Obsidian graph visualization

    line 78

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

Brainstorm -- LLM Knowledge Base Builder

Two modes: Ideation (think through problems) and Knowledge (build a persistent wiki from raw sources).

Architecture follows Karpathy's LLM Knowledge Bases pattern -- a structured, interlinked markdown wiki that compounds over time, browsable in Obsidian.

Quick Start

  • /brainstorm <topic> -- ideation mode (default)
  • /brainstorm ingest <source> -- knowledge mode, ingest a source
  • /brainstorm query <question> -- query the wiki
  • /brainstorm lint -- maintenance pass on the wiki

Architecture

See ARCHITECTURE.md for full wiki structure and Obsidian setup.

Ideation Mode

  1. Frame: "What are you solving? What's the constraint? What have you tried?"
  2. Diverge: Generate 5-10 approaches. Include one unconventional option. Do NOT evaluate yet.
  3. Converge: For each approach -- what breaks? what scales? what ships fastest? Connect to the student's production context.
  4. Crystallize: One direction, explicit next steps. Save to brainstorms/{topic}-{date}.md if worth persisting.

NEVER write code during brainstorming. Output is thinking, not implementation.

Discuss Mode (MANDATORY for every Q&A / brainstorm exchange)

Knowledge created from Principal-Partner conversation, not from an external source.

This mode fires automatically. It is NOT a judgment call. Every Q&A session, every brainstorm, every teaching exchange with durable conceptual content produces wiki artifacts in real-time alongside the knowledge/sessions/YYYY-MM-DD.md entry. The dual-artifact rule is specified in .claude/rules/session-capture.md.

Per exchange, the Partner does ALL of the following:

  1. Identify the atomic concept(s) the exchange surfaced (one concept per wiki page -- split if the exchange covered more than one).
  2. For each concept: write or update knowledge/wiki/{category}/{concept}.md using the template in ARCHITECTURE.md. If the page exists, merge new understanding in; do not overwrite the old.
  3. Add [[wiki-links]] to every related concept page that exists. Creating orphan pages is a failure -- every new node must connect to at least one existing node, or the page must explicitly flag itself as a new cluster root.
  4. Update knowledge/wiki/index.md with the new/modified entry under the correct category.
  5. Append a line to knowledge/wiki/log.md:
    ## [YYYY-MM-DD] discuss | {concept}
    - Added/Updated: wiki/{category}/{concept}.md
    - Linked to: [[page1]], [[page2]]
    - Source: knowledge/sessions/YYYY-MM-DD.md#exchange-N
  6. In the session exchange entry (knowledge/sessions/YYYY-MM-DD.md), the Concept linked: field points to the wiki page(s) created/updated this exchange. That is the bridge between the two artifact types.

Why this is mandatory, not optional: the wiki IS the lab's RAG corpus and the Principal's Obsidian second brain. Optional wiki creation = inconsistent corpus = RAG retrieval quality degrades = the research project fails silently. The Partner does not get to decide "this isn't worth a page." If a Q&A exchange happened, the concept earned its node.

Edge case -- trivial exchanges: if an exchange was pure clarification with no new concept (e.g., "what does this file path mean?"), no wiki page is needed. The test: would a future retrieval over this content surface useful signal? If no, skip the wiki write but still log the exchange in the session file.

Knowledge Mode (Ingest)

For external sources (articles, blog posts, documentation).

See ARCHITECTURE.md for the full ingest/query/lint workflows.

Quick version:

  1. Add source to knowledge/raw/
  2. LLM reads source, discusses takeaways with the student
  3. Writes summary page in knowledge/wiki/
  4. Updates knowledge/wiki/index.md
  5. Revises 5-10 related wiki pages with new cross-links
  6. Appends to knowledge/wiki/log.md

Rules

  • NEVER write code during brainstorming
  • Ask the student to explain their reasoning at every decision point
  • Challenge assumptions: "Why not the opposite approach?"
  • Connect to production reality and market relevance
  • If brainstorming reveals a knowledge gap, flag it for /review-deck
  • Use [[wiki links]] in knowledge files for Obsidian graph visualization

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.