agentleFS
QuarktexQuarktex/qcobrain/.claude/skills/learn/SKILL.md

learn

Skill4 starsChanged 4 months agoMITView on GitHub

Log a durable process or meta lesson — what happened, what to do differently, and why. Dual-writes a human-readable prose entry to runtime/learnings.md AND one structured JSON record to learnings/log.jsonl (same key, same schema), under the provenance guard — a lesson is durable only if it traces to a real, un-invalidated event. Enforces the anti-poison rule (no transient errors, no "X doesn't work" capability claims). A learning is about HOW TO OPERATE, never a fact about your business — facts go to wiki/ via /capture. Trigger on "log a learning", "note this lesson", "remember not to do that again", or a mistake/insight worth not repeating.

194 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use add-anydoc, add-clidash and 10 more.

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.
  • TomGranotSkill

    add-anydoc

    Add local office-document-to-Markdown conversion to NanoClaw agent containers with the pinned Firecrawl AnyDoc CLI. Use when agents need to read attached Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV, or text-based PDF files without uploading them to a hosted parser.

    Also in 5 projects that use this

  • TomGranotSkill

    add-clidash

    Add clidash — a zero-dependency, read-only web dashboard that derives its tabs and tables at runtime from any CLI that lists resources as JSON. Ships pre-wired for NanoClaw's ncl CLI (agent groups, sessions, channels, users, roles), plus message-activity charts, a log tail, and a read-only file viewer for group skills/CLAUDE.md/profiles.

    Also in 5 projects that use this

  • seek888Skill

    open-gstack-browser

    Launch GStack Browser — AI-controlled Chromium with the sidebar extension baked in. Opens a visible browser window where you can watch every action in real time. The sidebar shows a live activity feed and chat. Anti-bot stealth built in. Use when asked to "open gstack browser", "launch browser", "connect chrome", "open chrome", "real browser", "launch chrome", "side panel", or "control my browser". Voice triggers (speech-to-text aliases): "show me the browser".

    Also in 11 projects that use this

  • Saqlain-811Skill

    ios-sync

    Regenerate the iOS debug bridge against the latest upstream gstack templates. (gstack)

    Also in 6 projects that use this

  • MicrosoftSkill

    add-codex

    Use Codex (CLI + AppServer) as the full agent provider — planning, tool orchestration, native compaction, MCP tools, session resume — in place of the Claude Agent SDK. ChatGPT subscription or OPENAI_API_KEY. Per-group via agent_provider. Distinct from using OpenAI as an MCP tool (where Claude remains the planner).

    Also in 6 projects that use this

  • MicrosoftSkill

    add-deltachat

    Add DeltaChat channel integration via @deltachat/stdio-rpc-server. Native adapter — no Chat SDK bridge. Email-based messaging with end-to-end encryption.

    Also in 6 projects that use this

  • +5 more from TomGranot
  • +1 more from microsoft

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 169 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

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

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

    It is a learning, not a fact: If it is a claim about your business / market / people → stop and route it to /capture (cite-or-refuse into …

    line 26
  2. Very specific:

    Add an entry at the top (newest first), matching the file's format: a ## <YYYY-MM-DD> — <short title> header, the …

    line 42
  3. Very specific:

    Append-only: Never rewrite earlier lines or entries.

    line 69

“Common” is measured: kept by at least 6 in 10 of the 169 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

Inherits references/kernel.md (the provenance contract). Reads references/wiki-spec.md for the note schema and learnings/README.md for the learning-record schema.

What this is

qcobrain's "learnings log" under its provenance discipline. A learning is a lesson about how to operate — a pattern that worked, a pitfall to avoid, a preference you stated, a principle to follow. It is not a fact about you, your business, or the market — those are sourced into wiki/ via /capture. This skill is the sole writer of the learnings store, and it always writes both halves so every lesson is human-readable AND queryable.

The provenance guard (gate — do this FIRST, before writing anything)

  1. It traces to a real event. Name the actual occurrence the lesson came from — a real run, a real failure, a decision in decisions/log.md, a thing the user explicitly stated. No traceable event → do not record it as settled; at most log it source: inferred, confidence: low, ref: needs-source (a hypothesis, not a law).
  2. Anti-poison check (references/blocklist.md). REFUSE to log: transient / env-specific errors; negative tool-capability claims ("X doesn't work" — they harden into self-cited refusals); one-off chatter; secrets / PII; speculation stated as fact; the agent's own unverified beliefs. When in doubt, don't log — or log as inferred / low.
  3. It is a learning, not a fact. If it is a claim about your business / market / people → stop and route it to /capture (cite-or-refuse into wiki/), not here.

Execution

1 — Classify the lesson (per learnings/README.md)

  • type: pattern | pitfall | preference | principle | workflow
  • source (the provenance class): observed — you directly saw the triggering event (highest trust; the only class /skillify may promote into a skill) · user-stated — the user told you · inferred — synthesized from signals, treat as a hypothesis.
  • confidence: low | medium | high — independent of source (an inferred lesson can be high; an observed one can be low if it was a fluke).
  • key: a stable kebab slug naming the lesson. Reuse an existing key to supersede (newest wins); a genuinely new lesson gets a new key.

2 — Append the prose entry to runtime/learnings.md

Add an entry at the top (newest first), matching the file's format: a ## <YYYY-MM-DD> — <short title> header, the key/type/source/confidence line, the lesson (what happened / what to do differently / why), the triggering event, and what it traces to (ref). If you are superseding, mark the older prose entry superseded rather than deleting it (archive, never delete).

3 — Append the structured row to learnings/log.jsonl

Append one JSON object on a new line (append-only — never rewrite an earlier line), with the fields from learnings/README.md:

{"ts":"<ISO-8601 UTC, e.g. 2026-06-30T14:05:00Z>","key":"<kebab-slug>","type":"<pattern|pitfall|preference|principle|workflow>","source":"<observed|user-stated|inferred>","confidence":"<low|medium|high>","detail":"<one sentence: the lesson + what to do differently>","ref":"<sources key | wiki note id | decisions/log.md date | needs-source>","origin":"<user-directed|agent-generated>","status":"active"}

Set origin: agent-generated when a loop (e.g. /review) logged it, user-directed when the user did — the write-origin flag, so the loops only auto-manage what they wrote. To supersede, append a new line reusing the same key (newest wins); optionally re-append the prior line with "status":"superseded".

4 — Report

One line: the key, type, source/confidence, and confirmation that both halves were written (runtime/learnings.md + learnings/log.jsonl). If you refused, say which guard fired (anti-poison, or not-a-learning) and where the item should go instead (/capture, inbox/, or dropped).

Guardrails (hard)

  • No event, no fact. Untraceable lessons are inferred / low / ref: needs-source, never observed / high.
  • Both halves or neither. Prose to runtime/learnings.md AND a row to learnings/log.jsonl, kept in sync (same key, same ts). A learning that is only in one place is a bug.
  • Append-only. Never rewrite earlier lines or entries; supersede by reusing the key.
  • Anti-poison is non-negotiable. Transient errors and "X doesn't work" claims are dropped per references/blocklist.md — they are the exact poison the rule exists to keep out.
  • Learnings ≠ facts. A claim about your business belongs in wiki/ via /capture, not here.

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.