agentleFS
jeremylongshorejeremylongshore/bobs-big-brain-plugin/skills/brain/SKILL.md

brain

Skill2 starsChanged 4 months agoApache-2.0View on GitHub

Search and analyze governed notes, decisions, runbooks, and conventions, returning a qmd:// citation for every supported claim. Use when checking what the brain has captured about architecture, infrastructure, decisions, or operating practices. Trigger with "/brain", "ask the brain", or "check my knowledge base".

25 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use research.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • NVIDIASkill

    research

    Before designing a feature, research how comparable libraries solve the problem, check best practices, known pitfalls, and upstream changes. Use when starting a new feature, investigating an unfamiliar area, or the user asks to research something.

    Also in 4 projects that use this

The file

We split this skill into the rules it gives an agent

11 rules. 0 are common, 6 are distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 20 edited versions we read keep these.

  • 6

    Distinctive

    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:

    Ask your knowledge brain a question and get an answer grounded in your governed corpus, where every claim carries a qmd:// citation.

    line 21
  2. Very specific:

    Call brain_search with those keywords and scope: "curated" (default promoted knowledge).

    line 66
  3. Distinctive:

    Keyword retry (curated): only if the first call was still a long natural-language sentence (you skipped deriving 1–4 keywords).

    line 77
  4. Very specific:

    Scope retry (all): if curated is still empty, re-run the same keywords with scope: "all".

    line 80
  5. Distinctive:

    Only after curated+keywords and all+keywords are empty → Step 3 (honest refuse).

    line 82
  6. Distinctive:

    If two hits conflict, surface both with their citations rather than silently picking one — the governance layer tracks contradictions.

    line 95
  7. Very specific:

    If results is still empty after the empty-result ladder in Step 1, say so plainly: the brain has nothing governed on that topic.

    line 102
  8. Very specific:

    Read-only. This skill never writes to the corpus — capture and governance live in /brain-save.

    line 146
  9. Very specific:

    Never invent a qmd:// URI.

    line 148
  10. Distinctive:

    Prefer fewer, well-cited claims over a broad answer that cannot be anchored.

    line 149
  11. Distinctive:

    Do not conflate this product with IRSB / Moat / Scout (separate stack).

    line 150

“Common” is measured: kept by at least 6 in 10 of the 20 edited versions of this file we could read. “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

Brain — cited answers from your governed knowledge base

Ask your knowledge brain a question and get an answer grounded in your governed corpus, where every claim carries a qmd:// citation. The brain does not paraphrase from memory — it retrieves governed memories and cites them, so any answer is verifiable after the fact.

Overview

This is the read surface of Bob's Big Brain: your files are compiled into governed memories, governed by deterministic code, and retrieved with citations through qmd BM25, native FTS5 and configured dense retrieval. The brain_search MCP tool fronts that retrieval. Your product surround is INTKB govern + ICO compile + this plugin — not a fork of qmd.

The job here is to turn a natural-language question into a cited answer — and to refuse to answer beyond what the citations support.

Prerequisites

  • The bobs-big-brain / governed-brain plugin is installed (MCP governed-brain).
  • Mode is automatic. In local mode (default) search runs in-process against your local ~/.teamkb index — no remote brain service or API key; a compatible qmd binary must be available (prefer monorepo pin @tobilu/qmd via the Registrar / installer). Operators can use scripts/bbb-qmd from bobs-big-brain-registrar so XDG points at the team index, not personal ~/.cache/qmd. In team mode (TEAMKB_API_URL set) search proxies to the team brain. Every hit is a qmd:// citation.
  • See the runtime contract for exact inputs, result shapes, authentication, and mode-specific failures.

Instructions

Step 1: Search the governed corpus

The source entrypoints at the Registrar revision in gsb.lock.json configure Tobi's qmd BM25, native FTS5 and a sqlite-vec dense arm, joined with reciprocal-rank fusion and a freshness/category rerank. Dense is enabled by default in those entrypoints, but requires a built index and the configured loopback query embedder; disabling it or an unavailable arm leaves the lexical results. Team mode uses the deployed server's configuration. Retrieval makes no query-time generative LLM call.

Short keyword queries are useful for the lexical arms and empty-result retry; the configured dense arm also accepts natural-language phrasing. A zero-hit query is not proof that the brain lacks the fact.

  1. Derive 1–4 distinctive nouns/verbs from the user question (drop what/why/how, articles, auxiliaries, prepositions).
  2. Call brain_search with those keywords and scope: "curated" (default promoted knowledge).
brain_search({ query: "SOPS age secrets", scope: "curated" })

The tool returns { source, results: [{ citation, snippet, score, collection }] }. Each citation is a qmd://COLLECTION/FILENAME URI — the receipt for that hit.

Empty-result ladder (do not stop after one miss):

  1. Keyword retry (curated): only if the first call was still a long natural-language sentence (you skipped deriving 1–4 keywords). Re-run with 1–4 keywords only. Skip this step when the first query was already keyword-style — do not re-call with the same tokens.
  2. Scope retry (all): if curated is still empty, re-run the same keywords with scope: "all". Do not use inbox/archived unless the user asks.
  3. Only after curated+keywords and all+keywords are empty → Step 3 (honest refuse).
brain_search({ query: "shipped week", scope: "curated" })
// if empty (keywords already used):
brain_search({ query: "shipped week", scope: "all" })

Step 2: Answer ONLY from the cited results

  • Synthesize a direct answer from the returned snippets.
  • Attach the qmd:// citation to every claim, inline — for example: The proxy reverse-proxies the API (qmd://kb-curated/system-map.md).
  • If two hits conflict, surface both with their citations rather than silently picking one — the governance layer tracks contradictions; do not paper over them.
  • Do not add knowledge the citations do not support. Any reasoning beyond the corpus must be labeled clearly as inference, not the brain's answer.

Step 3: Handle an empty result honestly

If results is still empty after the empty-result ladder in Step 1, say so plainly: the brain has nothing governed on that topic. Do not fall back to general knowledge and present it as the brain's answer. Optionally note that the topic may need to be captured (run /brain-save).

Output

  1. A short, direct answer.
  2. Each load-bearing claim followed by its qmd:// citation.
  3. A closing Sources list of the distinct qmd:// URIs used.

Examples

Cited answer:

/brain what does my system map say about the proxy?

→ The proxy is the single ingress; it reverse-proxies each domain to its service
  and must be reloaded, not restarted, after edits (qmd://kb-curated/system-map.md).

Sources:
- qmd://kb-curated/system-map.md

Empty result (honest refusal):

/brain what is my refund policy?

→ The brain has nothing governed on a refund policy. I won't guess from general
  knowledge. If this should be in your brain, capture it with /brain-save.

Error Handling

SituationResponse
brain_search returns empty resultsRun the empty-result ladder (keywords → scope all); only then refuse. Do not fabricate.
qmd missing / index emptyRetrieval degrades to empty. Tell operator: install @tobilu/qmd, run INTKB pnpm search-canary -- --heal or bbb-qmd status.
MCP tool unavailablePlugin not enabled; install/enable governed-second-brain (governed-brain MCP server).
User asks to write/captureOut of scope here — direct them to /brain-save.

Guardrails

  • Read-only. This skill never writes to the corpus — capture and governance live in /brain-save.
  • Never invent a qmd:// URI. Cite only URIs returned by brain_search.
  • Prefer fewer, well-cited claims over a broad answer that cannot be anchored.
  • Do not conflate this product with IRSB / Moat / Scout (separate stack).
  • Provenance ≠ truth. Promotion receipts record each memory's capture channel (or unattested) and an origin-token hash — that proves WHERE a capture came from, not that its content is true. An authenticated insider can still have saved poisoned content with a valid origin; weigh cited answers accordingly and rely on govern policy, human review, and supersession as the content-quality gates.

Resources

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.