mirror of
https://github.com/alirezarezvani/claude-skills.git
synced 2026-10-09 03:17:54 +00:00
08740d4ec2
Applies every pre-merge required change from audit/pr-stream-2026-08/ that could not land on contributor forks, plus the stream's cross-cutting fixes: - stock-analysis (#944): description trimmed 1463 -> 1001 chars; Anti-Patterns + Cross-References sections added; security-auditor false positive at holdco-assetmgr.md:58 suppressed inline (auditor:ignore-line) -> strict PASS - deepread (#965): renamed research/dsh-deepread -> research/deepread; H1 and name updated; research-summarizer cross-ref path-qualified; 12 cited sources added across both references; plugin.json + marketplace entry; routed in the research orchestrator (SKILL.md registry + SIGNALS + classifier.py in lockstep, verified: 'deeply read this pdf' -> deepread, 3 signals) - business-name-fit (#926): 'Use whenever' -> 'Use when' (validator trigger regex); +2 cited sources (USPTO TMEP §1209, Usunier & Shaner 2002) -> 5 - embedded-iot-mentor (#942): references/hardware-selection.md (7 sources, datasheet-anchored) + worked mini-example; validator length gate now passes - swedish-mentor (#943): references/swedish-resources.md (6 sources, stable official URLs only); session recipes, milestones, learner situations, worked example; mandated opener softened to guidance; plugin.json + marketplace entry; validator length gate now passes - Related Projects (#895): LinkedIn Skills row trued up (10 -> 11 skills, hardcoded star count dropped) - check_plugin_json.py: marketplace description <= 1024 guard added to --all (the #964 regression guard; commercial-skills sits at 1021/1024) - #954 doc drift: quality_gates_for_skills.md, cs-skill-author.md, security-guidance SKILL.md now point attribution at authoring-notes.json - Counter true-up after the 6-skill merge batch: 370 skills / 672 tools / 809 refs / 92 plugins across README.md badge+table, CLAUDE.md, marketplace metadata (derive_counters.py --check passes) All gates green locally: frontmatter 0 errors, model freshness 0 findings, dual-publish 0 drifted, paths 0, smoke 0 failed, plugin-json 0 FAIL. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017Bzm6Pafyxja6g4jUDPcei
69 lines
3.5 KiB
Markdown
69 lines
3.5 KiB
Markdown
# Feynman Comprehension Loop
|
|
|
|
Use this reference for `feynman` mode. The objective is to expose and repair understanding gaps through retrieval, not to produce childlike wording for its own sake.
|
|
|
|
## Procedure
|
|
|
|
1. **Set the target** — name the idea, mechanism, or chapter-level claim to learn.
|
|
2. **Read for structure** — identify the central claim, key terms, mechanism, and evidence.
|
|
3. **Close the source** — do not look at the text or analysis notes.
|
|
4. **Explain from memory** — use plain language and preserve causal or logical links.
|
|
5. **Mark uncertainty** — tag every vague phrase, skipped step, undefined term, or appeal to jargon.
|
|
6. **Locate the first broken link** — find the earliest point where the explanation stops being justified.
|
|
7. **Return narrowly** — reread only the material needed to repair that gap.
|
|
8. **Rewrite the explanation** — replace jargon with a mechanism, concrete example, or boundary condition.
|
|
9. **Test transfer** — apply the idea to a new case and predict what should happen.
|
|
10. **Test contrast** — explain a case where the idea should not apply.
|
|
11. **Schedule retrieval** — create short recall prompts for later review.
|
|
|
|
## Gap Types
|
|
|
|
- **Definition gap** — a term cannot be explained without repeating itself.
|
|
- **Mechanism gap** — steps are listed but causality or logic is missing.
|
|
- **Evidence gap** — the claim is remembered but its support is not.
|
|
- **Boundary gap** — the learner cannot say when the claim fails.
|
|
- **Transfer gap** — the learner repeats the source but cannot use the idea elsewhere.
|
|
|
|
## Output Template
|
|
|
|
```markdown
|
|
## Learning target
|
|
[One precise concept or claim]
|
|
|
|
## Closed-book explanation
|
|
[Plain-language explanation reconstructed from memory]
|
|
|
|
## Detected gaps
|
|
- [Gap type]: [specific missing link]
|
|
|
|
## Corrected explanation
|
|
[Repaired explanation with the missing mechanism or evidence]
|
|
|
|
## Transfer test
|
|
[New scenario and predicted outcome]
|
|
|
|
## Boundary test
|
|
[Scenario where the idea should not apply]
|
|
|
|
## Recall prompts
|
|
1. [Question that retrieves the central mechanism]
|
|
2. [Question that retrieves evidence]
|
|
3. [Question that retrieves a limitation]
|
|
```
|
|
|
|
## Quality Rules
|
|
|
|
- Simplicity must not delete an essential condition.
|
|
- An analogy must identify where the analogy breaks.
|
|
- Reopening the source before attempting retrieval invalidates the gap test.
|
|
- Recognition (`this looks familiar`) is not recall (`I can reconstruct it`).
|
|
- Review questions should retrieve relationships and mechanisms, not isolated vocabulary.
|
|
|
|
## Sources
|
|
|
|
1. Adler, M. J. & Van Doren, C. — *How to Read a Book* (rev. ed., 1972) — the analytical-reading stages this workflow's deep mode follows.
|
|
2. Gleick, J. — *Genius: The Life and Science of Richard Feynman* (1992) — primary biographical account of Feynman's learn-by-teaching notebook practice.
|
|
3. Farnam Street — "The Feynman Technique: The Best Way to Learn Anything" (fs.blog/feynman-technique) — the canonical four-step formulation used here.
|
|
4. Karpicke, J. D. & Roediger, H. L. — "The Critical Importance of Retrieval for Learning", *Science* 319 (2008) — evidence that recall attempts, not re-reading, drive retention.
|
|
5. Dunlosky, J. et al. — "Improving Students' Learning With Effective Learning Techniques", *Psychological Science in the Public Interest* 14 (2013) — rates practice testing and self-explanation as highest-utility techniques, re-reading as low-utility.
|
|
6. Chi, M. T. H. et al. — "Self-Explanations: How Students Study and Use Examples", *Cognitive Science* 13 (1989) — the self-explanation effect behind the teach-back step.
|