claude-skill-writing-ecosystem

# claude-skill-writing-ecosystem

> A Claude Code skill bundle (Markdown only, MIT) that orchestrates the writing of **human-facing** articles, essays, blog posts and newsletter issues: one orchestrator skill, `writing-ecosystem`, takes a piece from a single central thesis agreed with the writer, through an editorial brief, drafting and a review panel, to the writer's content GO, title review and acceptance. **Bundles its six review agents** (editor / essay-reviewer / prose-clarity-reviewer / theme-reviewer / title-reviewer / fact-checker) so the skill and agents install together via `install.sh`. Audience-paired with `llms-txt-writer` (AI-facing documents).

Status: active; `skills/` and `agents/` are copied one way from the author's article repository [zenn-content](https://github.com/shimo4228/zenn-content) (`.claude/`), where they are used daily, so this copy can trail it between syncs.

## Core documentation

- [SKILL.md](skills/writing-ecosystem/SKILL.md): full skill specification — ecosystem map, canonical workflow (route, collect, editorial brief, draft, freeze, review panel, content GO, title, acceptance), finding-disposal rule, citation workflow, genre shapes, title conventions, project-overlay slot
- [references/](skills/writing-ecosystem/references/): style diagnostics (AI slop, JA + EN), review output format, publication procedures
- [README.md](README.md): human-facing narrative (requirements, install, ecosystem map)
- [CHANGELOG.md](CHANGELOG.md): release notes

## Bundled agents

- [editor](agents/editor.md): practical-channel article review — argument flow, explanation quality, AI slop, in-article terminology
- [essay-reviewer](agents/essay-reviewer.md): essay-channel review — logical structure, argument overload, tone
- [prose-clarity-reviewer](agents/prose-clarity-reviewer.md): first-read clarity checklist; used by the Codex plugin's `codex:codex-rescue`, or run directly as the fallback
- [theme-reviewer](agents/theme-reviewer.md): opt-in findings on a chosen question before writing; no verdict
- [title-reviewer](agents/title-reviewer.md): title-body contract after the content GO; findings only
- [fact-checker](agents/fact-checker.md): web verification of factual claims, local check of code / paths / output, paste-ready sources block

## Install

- [install.sh](install.sh): one-command bundler — `CLAUDE_HOME=<project>/.claude ./install.sh` copies skills/* → `<CLAUDE_HOME>/skills/`, agents/*.md → `<CLAUDE_HOME>/agents/`; idempotent, backs up replaced files to `<CLAUDE_HOME>/backups/install-<timestamp>/`, `--force` / `--dry-run`. Install per project: the agents read the skill from `<project>/.claude/skills/writing-ecosystem/`
- Requires a publication channel contract under `<project>/.claude/rules/` and a writing-principles file; example: [zenn-content/.claude/rules/](https://github.com/shimo4228/zenn-content/tree/main/.claude/rules)
- Named but not bundled (optional routes): `session-theme-mining`, `collect-context`, `headline-craft`, `quality-gate`, `prose-translation`

## The role-separation matrix

The skill enforces a role separation across the writing ecosystem:

|        | Write | Review: channel | Review: first read | Review: facts | Title review | Acceptance |
|---|---|---|---|---|---|---|
| **Component** | `writing-ecosystem` skill | `editor` / `essay-reviewer` agent | Codex `codex:codex-rescue` with the `prose-clarity-reviewer` checklist | `fact-checker` agent | `title-reviewer` agent | `quality-gate` skill (not bundled) |
| **Trigger** | first draft, full revision | after structural freeze | after structural freeze | before the writer's read-through | after the content GO | before the publication GO |

Which channel routes to `editor` or `essay-reviewer` is decided by the project's publication channel contract, not by this skill.

## Audience boundary

| Audience | Skill |
|---|---|
| Humans (articles, essays, blog posts) | `claude-skill-writing-ecosystem` (this one) |
| AI search engines and agents (`llms.txt`, FAQ) | `llms-txt-writer` |

The two skills do not overlap. Mixing audiences in one document makes both worse.

## Sibling skills

- [llms-txt-writer](https://github.com/shimo4228/llms-txt-writer): AI-facing documents (the audience-paired skill)
- [readme-writer](https://github.com/shimo4228/readme-writer): README counterpart, with a fresh judge checking the page against the code
- [claude-skill-paper-ecosystem](https://github.com/shimo4228/claude-skill-paper-ecosystem): academic papers (SSRN / arXiv / Zenodo) with five reviewer subagents
- [jsonld-knowledge-graph](https://github.com/shimo4228/jsonld-knowledge-graph): JSON-LD knowledge graph design for stable concept-level structure
- [context-sync](https://github.com/shimo4228/context-sync): project documentation audit
- [search-first](https://github.com/shimo4228/search-first): research-before-coding workflow
- [shimo4228](https://github.com/shimo4228/shimo4228): the author's hub — five practice lines with their DOIs, and the author's tools for Claude Code