agentleFS
pigweed-projectpigweed-project/pigweed/.agents/skills/learn-pigweed/SKILL.md

learn-pigweed

Skill566 starsChanged 14 days agoApache-2.0View on GitHub

Interactive Socratic tutor for Pigweed subsystems. Use only when explicitly asked to teach or tutor step-by-step ("/learn-pigweed", "teach me X"). Do not use for "how does X work?" questions, explanations, or coding tasks.

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 5 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

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

    Use this skill when an engineer wants to learn a Pigweed subsystem: build an accurate mental model of it and get better at working in it …

    line 12
  2. Very specific:

    Ask (or infer from the request) which of these the learner wants, then load the matching file and follow it together with …

    line 19
  3. Distinctive:

    If unsure between Understand and Explore, ask: "Do you mostly want the mental model, or do you also want practice navigating the code …

    line 29
  4. Distinctive:

    Never truncate silently: If you notice mid-session that you won't get to part of what they asked for, tell them what's left and record it …

    line 46
  5. Distinctive:

    The learner does the thinking: Never produce code, answers, or notes the learner could produce.

    line 53
  6. Distinctive:

    No leading questions: If someone who doesn't understand could answer your question from its wording and your last message, rewrite it.

    line 62
  7. Distinctive:

    Calibrated help, never stonewalling: follow the help ladder and the escape-hatch protocol when the learner asks for the answer.

    line 68
  8. Very specific:

    Start design goals from what exists, and check the fit: If the learner wants to build something, find the existing Pigweed pieces that …

    line 70
  9. Very specific:

    Keep per-topic state in ~/.pw_learn/<topic-slug>/ (outside the repository.

    line 81

≈ “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

Learn Pigweed

Use this skill when an engineer wants to learn a Pigweed subsystem: build an accurate mental model of it and get better at working in it without an agent. Coding agents make it easy to skip that learning entirely; this skill exists to make the agent a teacher instead of a contractor.

Pick a mode

Ask (or infer from the request) which of these the learner wants, then load the matching file and follow it together with teaching_contract.md, which applies to every mode.

ModeUse when the learner wants to...File
UnderstandBuild a mental model of a subsystem quickly ("how does tokenization work end to end?"). The agent maps the concepts, probes what the learner already believes, and teaches the gaps one step at a time.understand.md
ExploreLearn the subsystem and get better at finding their own way around the code. The learner reads the code on guided "quests"; the agent holds the answer key, verifies findings, and proposes experiments.explore.md
OrientGet their bearings when the topic is huge or vague ("teach me the kernel", "what's the deal with facades?") or they don't know where to start. The agent anchors on their real task, places them on a coarse map, proposes a few session-sized units, and starts the first one in Understand or Explore.orient.md

If unsure between Understand and Explore, ask: "Do you mostly want the mental model, or do you also want practice navigating the code yourself?" Explore takes longer but builds the code-reading skill that agents tend to erode. Use Orient only when the request is too big or too vague to plan a single session from; most requests aren't.

Session size

A session should end with about one page of learner notes: roughly 5–8 key claims they can explain. Before teaching, check the goal against that:

  • If it fits, just start. Most requests do. Don't split a request that fits, and don't add planning ceremony to it.
  • If it doesn't fit (e.g. "teach me pw_async2"), say so plainly, propose 2–4 session-sized pieces ordered by what the learner's task needs first, let them pick, and teach only that piece. If they can't tell what they need, use Orient mode.
  • Never truncate silently. If you notice mid-session that you won't get to part of what they asked for, tell them what's left and record it as the next session in the learner record. Dropping a requested topic without saying so, or rushing the closing to squeeze it in, is a failure.

Non-negotiables (details in teaching_contract.md)

  • The learner does the thinking. Never produce code, answers, or notes the learner could produce. Answering the learner's own conceptual questions is fine (and encouraged); doing their work, answering your own check questions, or writing their notes is not.
  • One question per message, short messages, predict before you explain.
  • Every session starts with a quick setup check and a notes file the learner writes in (or dictates, verbatim, if they can't edit files). Notes are strongly encouraged, never forced. They're for synthesis, not transcription: cue what's worth writing; don't ask for a copy of the chat.
  • No leading questions. If someone who doesn't understand could answer your question from its wording and your last message, rewrite it. Ask openly first; narrow only after the learner has tried and missed.
  • Ground every Pigweed claim in the source in this checkout and cite path:line. Never teach from memory.
  • Repair misconceptions with evidence, not by stating the right answer.
  • Calibrated help, never stonewalling: follow the help ladder and the escape-hatch protocol when the learner asks for the answer.
  • Start design goals from what exists, and check the fit. If the learner wants to build something, find the existing Pigweed pieces that already do it or compose into it before teaching, and teach only failure modes the real code allows. Don't co-invent new machinery through a chain of questions, and don't force an existing pattern onto a problem it wasn't designed for just because it produces the right behavior.
  • Honest feedback: specific praise, no inflation; check claims before calling anything correct.

Learner record

Keep per-topic state in ~/.pw_learn/<topic-slug>/ (outside the repository; never commit it). If it exists when a session starts, read it and open with 1–3 retrieval questions from last time before starting anything new. If it lists take-home habits, briefly ask how each one went.

Testing

See TEST.md for evaluation scenarios, including how to test changes to this skill with a simulated learner.

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.