From thinking-frameworks-skills
Ingests single inbox file into corpus/seeds/ as normalized markdown seed with full frontmatter. Orchestrates format normalization, topic tagging, intuition-density scoring, deduplication, changelog, ledger updates, and file move to .processed/. Useful for indexing raw notes, transcripts, or highlights.
npx claudepluginhub lyndonkl/claude --plugin thinking-frameworks-skillsThis skill uses the workspace's default tool permissions.
- [Workflow](#workflow)
Generates design tokens/docs from CSS/Tailwind/styled-components codebases, audits visual consistency across 10 dimensions, detects AI slop in UI.
Records polished WebM UI demo videos of web apps using Playwright with cursor overlay, natural pacing, and three-phase scripting. Activates for demo, walkthrough, screen recording, or tutorial requests.
Delivers idiomatic Kotlin patterns for null safety, immutability, sealed classes, coroutines, Flows, extensions, DSL builders, and Gradle DSL. Use when writing, reviewing, refactoring, or designing Kotlin code.
Related skills: Uses normalize-format, tag-by-topic, score-intuition-density, dedupe-against-corpus, update-topic-ledger. Called by the Librarian agent for each file in inbox/.
Copy this checklist and track progress per file:
Ingest one inbox file:
- [ ] Step 0: Compute sha256 of file body; check .librarian-state.json for duplicate fingerprint
- [ ] Step 1: Invoke normalize-format → body + partial frontmatter
- [ ] Step 2: Invoke tag-by-topic → topics + pending_tags
- [ ] Step 3: Invoke score-intuition-density → score + signals
- [ ] Step 4: Invoke dedupe-against-corpus → CREATE | LINK | SKIPPED
- [ ] Step 5: If LINK, backlink matched seeds' related_seeds field
- [ ] Step 6: Write seed file with full frontmatter; status=seed; manual_edits=false
- [ ] Step 7: Append one line to corpus/seeds/.changelog.md
- [ ] Step 8: Invoke update-topic-ledger for each topic tag
- [ ] Step 9: mv inbox file → inbox/.processed/
- [ ] Step 10: Update .librarian-state.json with the fingerprint
Step 0: Hash the body (not frontmatter). If the fingerprint is in .librarian-state.json, exit with SKIPPED (already ingested: <seed-id>). Idempotency is a feature.
Step 1: normalize-format handles all supported formats (plain markdown, .jsonl Claude Code sessions, .json Claude.ai exports, Readwise exports, transcripts with speaker labels, link captures). Returns one or more {body, partial_frontmatter} pairs — multi-chunk outputs possible for long transcripts.
Step 2: tag-by-topic proposes 1–4 tags from the controlled vocabulary. If none match, it logs to topic-ledger.md#pending-tags and still assigns the closest existing tag.
Step 3: score-intuition-density computes 0–10 from 8 explicit signals. Auditable.
Step 4: dedupe-against-corpus returns one of SKIPPED (exact fingerprint match — exit), LINK (near-match — proceed but write related_seeds), or CREATE (no match).
Step 6: Write the seed file. Location: corpus/seeds/{id}.md where id = YYYY-MM-DD-slugified-title. If a seed with that id already exists, append -v2 rather than overwrite.
Step 7: Changelog format: YYYY-MM-DDThh:mm | ADDED | {seed-id} | from {original_path} | density={N} | topics={comma-list}
---
id: 2026-04-23-dropout-as-ensemble-thinned-networks
title: "Dropout as ensemble over thinned networks"
created: 2026-04-21T09:14:00-07:00
source:
type: inbox-note
original_path: inbox/2026-04-21-dropout-as-ensemble.md
ingested_at: 2026-04-23T14:32:01-07:00
fingerprint: sha256:7c1d...
topics: [regularization, ensembling, dropout]
intuition_density:
score: 8
signals: [analogy_present, concrete_worked_example, counterfactual_offered, reframe_against_default, biology_to_ai]
status: seed
provenance:
author: kushal
confidence: owned
links:
related_seeds: [2026-03-11-l2-as-gaussian-prior]
parent_source: null
section_affinity: [agent-workshop]
word_count: 87
manual_edits: false
---
[body: preserved verbatim from normalize-format output]
Input: inbox/2026-04-21-dropout-as-ensemble.md (87 words, plain markdown, user's own note).
Run:
[regularization, ensembling, dropout]; dropout new → logged to pending-tags.2026-03-11-l2-as-gaussian-prior at Jaccard 0.32 (below 0.5 threshold) → CREATE (not LINK).corpus/seeds/2026-04-21-dropout-as-ensemble-thinned-networks.md with full frontmatter.2026-04-23T14:32 | ADDED | 2026-04-21-dropout-as-ensemble-thinned-networks | from inbox/2026-04-21-dropout-as-ensemble.md | density=8 | topics=regularization,ensembling,dropoutregularization seeds 3→4, temperature→hot; ensembling 1→2; dropout new row added.mv inbox/2026-04-21-dropout-as-ensemble.md inbox/.processed/Output: one seed file, one changelog line, three ledger updates, one moved file.
normalize-format fails, the inbox file stays put (not moved) and the changelog records ERROR | <file> | <reason>. No partial ingests.id exists, append -v2; never overwrite.inbox/.processed/ or inbox/.trash/.dedupe-against-corpus returns SKIPPED, move the inbox file to .processed/ with a renamed suffix -duplicate-of-{matched-seed-id} and exit — don't re-write the matched seed.