agentleFS
Corvus400Corvus400/notebooklm-study-kit/.claude/skills/regenerate-playbook/SKILL.md

regenerate-playbook

Skill2 starsChanged 5 months agoMITView on GitHub

Regenerate ONLY the playbook of an EXISTING topic in this NotebookLM corpus monorepo for a different learner level — without re-collecting or touching the corpus. Use when the learner has leveled up (or wants an easier/harder pass) and the existing playbook no longer fits. Trigger on 「playbook だけ作り直して」「中級者向けに(playbook を)再生成」「レベルが上がったので playbook を更新」 「上級者向けに playbook を作り直して」, "regenerate the playbook", "rebuild the playbook for a different level", "redo just the playbook for <topic>". Make sure to use this skill whenever the user wants a fresh playbook for a topic they ALREADY have a corpus for, even if they don't say the word "regenerate". Do NOT use this for a brand-new topic with no corpus yet (that is `add-topic`), for collecting or editing corpus files, for running a playbook step (self-test etc.), or for fixing the tools — those are different tasks.

  • In this repositoryOne other file it gives its agents.

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    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. Very specific:

    Use AskUserQuestion (options in the user's language) to pick one of the five learner paths defined in METHOD.md "Learner paths" — this is …

    line 73
  2. Very specific:

    If that helper does not exist yet, do the equivalent by hand: copy playbook-template/*.md (the chosen language, stripping the .ja marker) …

    line 97
  3. Very specific:

    Variant selection (delete, don't keep both): keep the 03-tensions.md and 04-trajectory.md variant that matches the recovered archetype + …

    line 111
  4. Very specific:

    Read topics/android/playbook/ as the reference depth.

    line 122
  5. Very specific:

    No placeholders left; exactly one variant in 02/03.

    line 142

≈ “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

regenerate-playbook

Rebuilds the playbook only for an existing topic under topics/<slug>/, retargeted to a chosen learner path, while leaving the collected corpus completely untouched. The expensive, judgement-heavy work in this kit is corpus collection (add-topic Steps 4–6); the playbook on top of it is cheap to redo. As a learner moves from complete beginner to senior, the same corpus should be driven by a different playbook — that is exactly what this skill produces.

The learning method and the learner paths live in METHOD.md. Read its "Learner paths" section before customizing — this skill orchestrates the regeneration but does not restate the path definitions (single source of truth, shared with add-topic).

Boundary with add-topic (read this first)

add-topic is for a new subject with no corpus yet; its own description says it must not fire for work inside an existing topic. regenerate-playbook is the opposite: the topic already exists (topics/<slug>/corpus-config.json, corpus/, coverage-map.md are present) and the corpus stays as-is. If the user actually wants a new subject, hand off to add-topic instead.

How to work: interactive, with checkpoints

Retargeting a playbook is a judgement call (which learner path, which variant). Do not run end to end autonomously — confirm the target topic, the recovered values, and the chosen path with the user before rewriting. Never overwrite corpus/, coverage-map.md, review-log.md, or sources/ — those are the learner's accumulated state.

Depend only on the shared layer (METHOD.md / playbook-template/ / tools/). Treat topics/android/playbook/ as a worked example to imitate if present, never a hard dependency.

Step 1 — Identify the target topic

Confirm which existing topic to regenerate. A valid target has topics/<slug>/corpus-config.json and a non-empty coverage-map.md (the corpus may be present locally or excluded from the public repo — either is fine; you do not read corpus content here). If no such topic exists, stop and point the user at add-topic.

Step 2 — Recover only what is actually persisted

The scaffold deliberately stores very little, so be honest about what can and cannot be recovered:

  • NOT persisted — corpus-config.json holds only required_frontmatter_keys / min_words / max_words / failure_markers (tools/scaffold_topic.py write_corpus_config). It does not store the domain archetype, the source policy, or the sub-domain list. Those were chosen via AskUserQuestion during add-topic and consumed by the editorial pass.
  • Recoverable — derive the topic's themes / sub-domains from the theme table in coverage-map.md (| NN | <theme> | … | rows). Read {TOPIC} and {ROLE} off the existing playbook's headings / intro. Infer the archetype + source policy from which variant currently survives in 03-tensions.md (A = disagreement-proxy / B = real disagreements) and 04-trajectory.md (A deprecation / B interpretations / C blueprint diffs), and present that as the default.
  • Re-ask the archetype + source policy with AskUserQuestion (options in the user's language, default = the inferred one) so the regenerated variants match reality. See METHOD.md §4.5 / the add-topic Step 2 row mapping.

Tell the user the recovered values and the inferred defaults before proceeding.

Step 3 — Choose the learner path (ask the user)

Use AskUserQuestion (options in the user's language) to pick one of the five learner paths defined in METHOD.md "Learner paths" — this is the whole point of the regeneration:

  1. Complete beginner — barely knows the field; fixed starter source subset, easiest generate-first, 07 self-test heaviest (06 diagnosis optional), tracks progress with NotebookLM native Notes/Flashcards (not review-log.md).
  2. Intermediate, blind to gaps — runs a brisk 02–05 for context, then 06 weak-spot diagnosis, then 07/08 on the weak themes.
  3. Weak-point / pre-interview — skims 02–05, uses 06 to confirm weak themes, concentrates on 07/08, finishes with 09.
  4. Review / long-term retention — 08-centric spaced review; 07 only for weak items.
  5. Senior aiming for expert — deepest sources (tensions/edge-cases/primary specs), hardest generate-first (predict where experts disagree), 03 + 06 + expert-depth 09.

The chosen path becomes the default path highlighted in the regenerated 00-how-to-use.md's "choose your path" section; the other paths remain as alternatives.

Step 4 — Regenerate the playbook from the template (corpus untouched)

Copy the playbook from playbook-template/ for the topic's language, substituting {TOPIC}/{ROLE} and the title marker — the same mechanical step add-topic Step 3 runs, but playbook files only.

python3 tools/regenerate_playbook.py --slug <slug> --topic "<TOPIC>" \
    --role "<ROLE>" --lang <en|ja> [--force]

If that helper does not exist yet, do the equivalent by hand: copy playbook-template/*.md (the chosen language, stripping the .ja marker) into topics/<slug>/playbook/, run the same {TOPIC} / {ROLE} / (Template) / (テンプレート) substitution as tools/scaffold_topic.py:substitute, and leave corpus/, corpus-config.json, coverage-map.md, review-log.md, sources/ exactly as they are. Because this overwrites existing playbook files, confirm with the user first (or take a quick git snapshot) so a half-finished regeneration is never silently shipped.

Step 5 — Editorial pass keyed to the chosen path

The scaffold only token-substituted. Apply the same editorial customization as add-topic Step 8, biased toward the chosen learner path:

  • 00-how-to-use.md: set the chosen path as the recommended default in the "choose your path" section, and specialize scope/aim and the per-prompt recommended source subsets to that path.
  • Variant selection (delete, don't keep both): keep the 03-tensions.md and 04-trajectory.md variant that matches the recovered archetype + policy; delete the others.
  • 02-mental-models.md: replace the domain-variant placeholders ({domain 1..4} / {領域 1..4}) with the topic's actual sub-areas from the recovered theme list.
  • 03-tensions.md: re-seed the tension axes (reuse the existing coverage-map.md Confusable-concepts pairs; do not regenerate that section — it is the learner's, not yours).
  • 04-trajectory.md / 05-interview-questions.md / 99-corpus-audit.md: specialize the scope lines.
  • Generate-first difficulty and step emphasis follow the path table — e.g. soften the "predict where opinions split" prompt for the complete-beginner path, sharpen it for the expert path.

The topic is single-language, so each prompt is one plain NN-*.md — apply the customization to it. Read topics/android/playbook/ as the reference depth.

Step 6 — Validate

python3 tools/validate_corpus.py topics/<slug>

Fix what it reports and re-run until PASS (or, if the corpus is excluded locally, confirm it still parses the coverage-map). Confirm directly that:

  • No {TOPIC} / {ROLE} / (Template) / (テンプレート) remain in topics/<slug>/playbook/.
  • Exactly one variant remains in 03-tensions.md and 04-trajectory.md.
  • corpus/, coverage-map.md, review-log.md, sources/ are unchanged (git status shows only playbook/ files touched).

Definition of done

  • topics/<slug>/playbook/ is regenerated and customized for the chosen learner path, with that path set as the default in 00-how-to-use.md.
  • No placeholders left; exactly one variant in 02/03.
  • corpus/, corpus-config.json, coverage-map.md, review-log.md, sources/ are untouched (verified via git status).
  • The user knows which path was applied and that they can re-run this skill to retarget again as they level up.

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.