writing
CLAUDE.md3 starsChanged yesterdayCC0-1.0View on GitHub
This repository contains published, author-voiced Zenn / Dev.to articles and note / Substack essays about AI agents, Claude Code workflows, and LLM engineering. Study drafts without an author voice do not belong. The repository is public; nothing load-bearing depends on a publishing platform remaining available. README is a routing page, not an article list. The exhaustive generated index is docs/PUBLICATIONS.md. Human-primary prose has one entrypoint, resident in this repository: .claude/skills/writing-ecosystem/SKILL.md (the review agents review-structure / review-certainty / review-explanation / review-prose…
In this repository12 other files it gives its agents.
- llms-full.txtllms.txt
- llms.txtllms.txt
- article-interviewSkill
- article-stocktakeSkill
- cover-artSkill
- note-publishingSkill
- publish-articleSkill
- quality-gateSkill
- session-theme-miningSkill
- substack-publishingSkill
- writing-ecosystemSkill
- zenn-formatSkill
The file
We split this CLAUDE.md into the rules it gives an agent
5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
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
- could be automatic: a setting could enforce it
- Very specific: line 15
verify drift before handoff: npm run check:index
- Distinctive: line 16
do not hand-maintain article titles or counts in README / llms files
- Very specific: line 29
Do not restate the backbone or the shared flow here.
- Very specific: line 75
If article, schedule.json, or generated-index changes are committed, remind the user to push.
- Distinctive: line 80
Never commit personal file paths, API keys, credentials, unsanitized raw logs, or sensitive screenshots.
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md 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.
Files
Claude Code Instructions for the writing repo
Project Overview
This repository contains published, author-voiced Zenn / Dev.to articles and note / Substack essays about AI agents, Claude Code workflows, and LLM engineering. Study drafts without an author voice do not belong. The repository is public; nothing load-bearing depends on a publishing platform remaining available.
Publications index
README is a routing page, not an article list. The exhaustive generated index is docs/PUBLICATIONS.md.
- source:
articles/*.mdfrontmatter,scripts/schedule.json,scripts/corpus.yml,scripts/reading_paths.yml - generate after article / URL / essay / paper changes:
npm run generate:index - verify drift before handoff:
npm run check:index - do not hand-maintain article titles or counts in README / llms files
Writing harness
Human-primary prose has one entrypoint, resident in this repository:
.claude/skills/writing-ecosystem/SKILL.md (the
review agents review-structure / review-certainty / review-explanation / review-prose / review-conventions / prose-clarity-reviewer /
fact-checker and the skills quality-gate / session-theme-mining live in .claude/ too).
The author's writing values and principles live only in .claude/rules/writing-principles.md (the backbone).
Platform and repository values live only in .claude/rules/publishing-channels.md.
For article work in another repository, add this repo with claude --add-dir ~/MyAI_Lab/writing
or start the session here.
Do not restate the backbone or the shared flow here. writing-ecosystem owns the procedure: the editorial
brief, review order, revision return conditions, and the boundary that excludes ADRs / memory / raw session
history from writing-time rules. Article direction comes from interviewing the author
(.claude/skills/article-interview/); outlines and plans for an article are written only after its brief.
The project output style .claude/output-styles/zenn-writing.md (selected in .claude/settings.json) governs how
Claude talks with the author, not article prose. For coding work such as scripts/, set "outputStyle": "default" in
.claude/settings.local.json (CLI: /output-style default).
Channel and format boundaries
- Zenn:
articles/; format and syntax are owned byzenn-format - Dev.to:
articles-en/; translation method is globalprose-translation, platform conversion isdevto-translator - note:
note/(note 初出の JA 正本と、Zenn 記事の原文転載。転載の正本はarticles/) - Substack EN translation:
substack/(frontmatter の有無は publishing-channels.md の Deterministic checks 列が持つ) - note publishing and corpus mirroring:
note-publishing - Substack publishing and corpus mirroring:
substack-publishing - note daily queue and scheduling:
docs/note-pipeline.md - Zenn / Dev.to publishing pipeline:
publish-article
Exact register, audience promise, review aspects, title limit, published_at, cadence, terminology, and
publish handoff are defined in .claude/rules/publishing-channels.md; do not duplicate their values here.
Drafting stays in the orchestrator. Review and acceptance follow writing-ecosystem (resident in
.claude/skills/) plus the local channel contract.
Post-publication measurement
article-stocktake is project-local instrumentation. It collects and reports reception differences, then
stops. Human-side reception data lives in scripts/metrics/snapshots.jsonl: Zenn views and note
impressions / PV / likes are read from each platform's dashboard (source: zenn_dash / note_dash, one row
per article plus an account row), alongside the older Zenn / Dev.to API rows. Check this file before
claiming human reception is unmeasured; the row format and collection procedure belong to article-stocktake. It does not generate or rank themes and does not call session-theme-mining automatically.
Verification and publishing
Pipeline details live in the scripts (scripts/devto_crosspost.py --help, scripts/note_publish.py --help) and the
publishing skills below; the schedule.json schema is .claude/refs/schedule-schema.md.
/quality-gate <file>- author publication GO
/publish-article <file>for Zenn / Dev.to,note-publishingfor note, orsubstack-publishingfor Substacknpm run generate:indexnpm run validateandnpm run check:index
If article, schedule.json, or generated-index changes are committed, remind the user to push. An unpushed
commit does not reach Zenn scheduling or the Dev.to cross-post job.
Public-repository safety
Never commit personal file paths, API keys, credentials, unsanitized raw logs, or sensitive screenshots.
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
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.

