agentleFS
paslavskyipaslavskyi/llm-wiki/.claude/skills/kb-elicit/SKILL.md

kb-elicit

Skill1 starsChanged 33 days agoMITView on GitHub

Use to run a structured knowledge-gathering dialogue in this repo — collaboratively frame the mind-map of areas, then drill top-down, turning the conversation into atomic notes. Reusable across knowledge bases via kb.framework.yml.

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 2 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read kb.config.yml (mode, language) and run kb-orient.

    line 14
  2. Very specific:

    Read kb.framework.yml. Propose its areas as a recommended top-level skeleton — explicitly a starting point, not a mandate.

    line 16
  3. Very specific:

    Update STATE.md: map framed.

    line 22
  4. Very specific:

    Update STATE.md (current focus, next step — intent only.

    line 43
  5. Very specific:

    Run npm run clarify -- [--topic TOP-xxx]: the ranked queue of open Q- notes and inline markers (must-priority, referenced and blocking …

    line 55
  6. Distinctive:

    Ask at most 5 questions per round, one at a time, in queue order.

    line 57
  7. Distinctive:

    After EACH answer, immediately write it through the pipeline:

    line 60
  8. Very specific:

    Stop after 5, or when the owner says stop / defers.

    line 66
  9. Very specific:

    Never edit index/ by hand.

    line 85

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

kb-elicit — dialogue engine (mind-map)

Goal: turn a conversation into a well-structured knowledge graph, top-down, saving each note only once it is a logically consistent, complete thought.

Method

Stage 0 — frame the map (when the map is empty, or on request)

  1. Read kb.config.yml (mode, language) and run kb-orient. Check whether a map exists (any type: topic notes / index/mindmap.md).
  2. Read kb.framework.yml. Propose its areas as a recommended top-level skeleton — explicitly a starting point, not a mandate.
  3. Adapt WITH the user: add / rename / drop areas to fit this base.
  4. For each agreed top-level area, create a topic note (parent: null) via kb-capture, placed in knowledge/<key>/. Use the area's framing questions to seed the conversation.
  5. Update STATE.md: map framed; recommended next area. (Intent only — never copy counts/progress; those are generated in index/. See CLAUDE.md → "STATE.md scope".)

Stage 1 — drill top-down (when a map exists)

  1. Present the agenda (see "Coverage / agenda" below) and propose where to dive. Agree on a node with the user.
  2. Frame the node: propose "what's worth collecting here" (sub-aspects). Capture agreed sub-aspects as child topic notes (parent: <node>). A node's skeleton lives as prose in its body + child topics — there is no rigid checklist field (LLM-native: structure is the graph + prose).
  3. Fill each sub-aspect: converse, asking clarifying questions, UNTIL the note is logically consistent (a complete, coherent thought). Only then save it via kb-capture with topic: <node>.
  4. Anything unresolved is recorded, not dropped. A gap local to ONE note stays inline in that note: [NEEDS CLARIFICATION: …] / [ASSUMPTION: …] (the note remains draft/proposed). A gap that spans notes or needs its own decision becomes a note: open question -> Q-, assumption -> ASMP-, risk -> RISK-, attached to the same node.
  5. On request, propose options or research (web/evidence) and fold conclusions back into notes — every external finding becomes a source note (SRC-) cited via sources: (see kb-capture → Evidence check).
  6. Update STATE.md (current focus, next step — intent only; no counts/progress, see "STATE.md scope").

Saving rule

Save when a note is logically consistent. Ask a clarifying question instead of saving a half-formed note. Confirm with the user only when a write OVERWRITES existing knowledge; new, consistent knowledge is captured as it forms.

Recurring concern? If the same risk/assumption/trade-off recurs across ≥2 notes, canonicalize it into ONE note and link the others to it — don't re-state the same thought in multiple bodies (see kb-capture "Recurring-concern check").

Clarify loop (resolve gaps — on request, or when the agenda is mostly gaps)

  1. Run npm run clarify -- [--topic TOP-xxx]: the ranked queue of open Q- notes and inline markers (must-priority, referenced and blocking items first).
  2. Ask at most 5 questions per round, one at a time, in queue order. Each question offers a recommended answer (with a one-line why) plus 2–4 alternatives, or asks for a short free answer. Never batch them into a wall.
  3. After EACH answer, immediately write it through the pipeline:
    • edit the affected note: replace the marker / contradicting text with the answer (a resolved Q- → status: accepted with the answer in its body, linked via relations: { answers: [Q-…] } from the note it settles);
    • append to that note's body, under ## Clarifications → ### YYYY-MM-DD: - Q: <question> → A: <answer> (append-only; the dated log of why it changed).
  4. Stop after 5, or when the owner says stop / defers. Deferred items stay as markers / open Q- — never silently dropped. Report: resolved / deferred / still open.

Coverage / agenda (what to work on next)

Compose, do not maintain a separate drifting artifact:

  1. Empty areas — from index/MAP.md (count 0).
  2. Open Q- / ASMP- / RISK- notes and "Unresolved markers" in index/health.md — known gaps.
  3. Topic nodes with no attached notes — from index/mindmap.md.
  4. Coverage gaps — from index/coverage.md (e.g. a must-requirement no story delivers, a pain nothing addresses, a risk nothing mitigates).

Output (per mode)

  • debug: state which notes you created/updated, their ids, and the agenda.
  • autonomous: short human acknowledgements; hide ids and wiki mechanics.

Rules

  • Capture only via kb-capture; resolve-before-write to avoid duplicates.
  • Never edit index/ by hand. After writes, the Stop hook reindexes.
  • One concept per note. Keep topic/parent pointing at real TOP- notes.
  • Suggest running kb-visualize when the user wants to see the map.

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.