agentleFS
DeupaxxDeupaxx/EveryDay-Writer/CLAUDE.md

EveryDay-Writer

CLAUDE.md39 starsChanged 56 days agoMITView on GitHub

Everyday Writer (EW) is a Claude Code skill plugin for professional writers. It has one root dispatcher, 13 writing sub-skills, voice management, and local workspace management, all constrained by a voice fingerprint and anti-AI writing rules. Invoked via /ew:ew or any of the sub-skill commands. The system holds any number of voices — the writer's own, plus one per ghostwriting client — with one active at a time. Plugin identity is defined in .claude-plugin/plugin.json (name: ew). These are not style…

The file

We split this CLAUDE.md into the rules it gives an agent

6 rules. About 0 are common, 1 is distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    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:

    Run npm test, npm run check, and npm run check:claude-skill before any commit that touches installation, manifests, skill frontmatter, …

    line 22
  2. Very specific:

    Every /ew invocation in SKILL.md follows this sequence:

    line 34
  3. Distinctive:

    Every skill must open with YAML frontmatter, then an H1, then its dependency chain block naming the runtime, workspace, anti-AI, …

    line 93
  4. Very specific:

    When adding a section to anti-ai-rules.md, check whether the skills' dependency-chain blocks or SKILL.md's standards section reference …

    line 96
  5. Very specific:

    When a core rule changes, it must be updated in both core/anti-ai-rules.md and cowork/ew-master.md.

    line 102
  6. Very specific:

    All registered sub-skills name both core/obsidian-workspace.md and core/voice-profile.md in their dependency chains, and check.js …

    line 108

≈ “Common” is an estimate: at least 1 in 5 of the 27,641 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

What This Project Is

Everyday Writer (EW) is a Claude Code skill plugin for professional writers. It has one root dispatcher, 13 writing sub-skills, voice management, and local workspace management, all constrained by a voice fingerprint and anti-AI writing rules. Invoked via /ew:ew or any of the sub-skill commands.

The system holds any number of voices — the writer's own, plus one per ghostwriting client — with one active at a time.

Plugin identity is defined in .claude-plugin/plugin.json (name: ew).

Hard Requirements for Plugin Discovery

These are not style preferences. Getting any of them wrong makes skills silently fail to register, with no error message anywhere:

  1. Every skill file needs YAML frontmatter with name and description. This applies to root SKILL.md and all skills/*/SKILL.md. A file without it is invisible to the loader.
  2. The manifest lives at .claude-plugin/plugin.json. A root-level plugin.json is ignored.
  3. There is no entry field in the manifest schema. Skills are discovered by directory convention: root SKILL.md plus each skills/<name>/SKILL.md.
  4. Skill names must be lowercase kebab-case and should match their directory name.

Run npm test, npm run check, and npm run check:claude-skill before any commit that touches installation, manifests, skill frontmatter, dependency chains, reference loading, or package contents. Installer tests use isolated temporary homes and vaults; never point them at a real user vault.

Entry Points

  • SKILL.md — master dispatcher, registered as skill ew
  • onboarding/ONBOARDING.md — environment and optional-storage router, then voice router (own voice vs subject)
  • bin/install.js — Node 16 installer, optional --vault bootstrap connection, and uninstall
  • .claude-plugin/plugin.json — plugin manifest
  • .claude-plugin/marketplace.json — marketplace manifest

Architecture: 4-Step Dispatch Flow

Every /ew invocation in SKILL.md follows this sequence:

  1. Workspace Check — resolve default storage or the optional connected Obsidian workspace through core/obsidian-workspace.md. Status/connect/disconnect/migrate/topic repair, voice repair, and world-builder journal repair bypass normal resolution for command-local recovery validation.
  2. Profile Check — resolve the active profile through core/voice-profile.md; if no voice exists, trigger onboarding.
  3. Onboarding — route through onboarding/ONBOARDING.md → either claude-code-mode.md (7-step, file-based) or claude-ai-cowork-mode.md (memory-based).
  4. References Check and dispatch — load active-voice references, then the sub-skill. Direct writing skills read runtime → workspace → anti-AI core → voice resolver → references → sub-skill.

Direct invocation (/ew:linkedin) bypasses step 4's routing but not steps 1–3 or the read order.

Core Constraints (read before every write)

FilePurpose
core/runtime-contract.mdShort operating contract: read order, precedence, fabrication rule, two-pass loop, final checklist.
core/obsidian-workspace.mdLocal data-root resolution, containment, topic layout, filenames, persistence, and graph links.
core/anti-ai-rules.mdThe primary standard. Sections 0–10.
core/ai_slop_commandments.mdTechnical companion: mechanism behind each failure pattern, era-indexed slop vocabulary, diagnostic checklist.
core/voice-profile.mdThe resolver, not a profile. Points at the active voice.
core/voice-profile-template.mdBlank fingerprint each new voice is stamped from.

Key sections of anti-ai-rules.md:

  • §0 — the operating contract (A-Player standard)
  • §0.1 — precedence. Voice profile outranks style rules; a pasted sample outranks the profile. Nothing overrides §0 or §0.2.
  • §0.2 — the fabrication rule. Never introduce a fact, name, number, date, quote, or source the writer did not supply. Use visible [brackets] to mark gaps.
  • §1–5 — banned words, sentence patterns, tone patterns, structural patterns, formatting tells
  • §6 — craft rules (King, Lamott, Brande)
  • §7.1 — the two-pass loop. Never present a first draft.
  • §7.2 — pre-publish checklist
  • §8 — pointer to the voice profile
  • §9 — what NOT to flag. False positives; count clusters, not instances.
  • §10 — signs of human writing to preserve.

§9 and §10 are load-bearing. Without them the banlist over-corrects and strips whatever made the prose a person's.

Skills Directory

skills/
  newsletter-creative/SKILL.md   # 5-beat Hero's Journey structure
  newsletter-technical/SKILL.md  # Tutorial/analysis format
  linkedin/SKILL.md              # Posts and articles
  tweets/SKILL.md                # Twitter/X posts and threads
  substack-notes/SKILL.md        # Substack Notes format
  web-copy/SKILL.md              # Landing pages
  sales-copy/SKILL.md            # Direct response, sales pages
  scene-structure/SKILL.md       # Scene construction
  script-writing/SKILL.md        # Screenplays
  world-builder/SKILL.md         # Journaled multi-file world-building and operation repair
  audit/SKILL.md                 # Before/after rewrite comparison
  outline/SKILL.md               # Idea-to-outline engine
  failure-library/SKILL.md       # Annotated AI-slop failure patterns
  voice/SKILL.md                 # Voice management: list, switch, new, import, edit, delete, repair
  workspace/SKILL.md             # Obsidian connection, migration, status, topic projects, repair

Developing Skills

  • Skills are Markdown instruction files. Installer and packaging code is dependency-free Node 16 CommonJS. npm test covers installer behavior, npm run check validates the instruction graph, and npm run check:claude-skill validates the ZIP boundary.
  • Every skill must open with YAML frontmatter, then an H1, then its dependency chain block naming the runtime, workspace, anti-AI, slop-pattern, and voice contracts. This is non-negotiable — direct invocation bypasses the dispatcher, so a skill that omits a core file simply never reads it.
  • The root references/ folder is deprecated and must not be scanned. User reference files live at <data-root>/voices/<slug>/references/, resolved by the workspace and voice contracts; only references/DROP-MD-FILES-HERE.md belongs in the root folder by default.
  • resources/ holds background reference material (writing commandments, rules rundowns), not part of the dispatch system.
  • When adding a section to anti-ai-rules.md, check whether the skills' dependency-chain blocks or SKILL.md's standards section reference section numbers that need updating.

The Cowork Port

cowork/ is a parallel packaging of the same system for Claude.ai Projects, which cannot load file-based plugins. Files there are self-contained and carry their own frontmatter.

When a core rule changes, it must be updated in both core/anti-ai-rules.md and cowork/ew-master.md. The cowork files are copies, not references. Nothing keeps them in sync automatically.

Multi-Voice Architecture

core/voice-profile.md is a resolver, not a profile. Anyone opening it expecting a fingerprint will be confused, so this is the most important non-obvious fact in the repo.

All registered sub-skills name both core/obsidian-workspace.md and core/voice-profile.md in their dependency chains, and check.js hard-errors when either contract disappears. Do not bypass these resolvers with literal storage paths.

Voices live outside the plugin under the resolved data root. Default mode uses the bootstrap tree below; connected mode uses the same voice tree under <vault>/Everyday Writer/ and adds a human-readable home note plus topic projects.

The installer accepts --vault <existing-vault>, or asks only when both stdin and stdout are TTYs and bootstrap storage is safe and empty. It may configure only a new/empty fixed direct child named Everyday Writer; existing bootstrap or destination data belongs to the runtime migration flow. config.json is published atomically with no-clobber semantics only after canonical vault and workspace validation. A plain install/update must continue with a target-specific recovery warning when storage configuration is broken; explicit --vault remains a hard stop. Package trees are staged, ownership-receipted, locked, quarantined, and atomically swapped so updates cannot overlay stale files. Uninstall must require the matching receipt and must never remove configuration, legacy voices, or vault knowledge.

~/.everyday-writer/
  active-voice                  one line: the active slug
  voices/<slug>/
    voice-profile.md            the actual fingerprint
    references/                 per-voice drop-zone
    drafts/                     file-mode output
    samples/                    provenance

Two reasons they cannot live in core/: the plugin directory is a cache that updates replace wholesale, and this repo is public. A client profile in core/ would be destroyed on upgrade and one commit from publication. check.js fails the build if a completed profile appears there.

Resolution rules, every failure case, and the inline-override behavior are in core/voice-profile.md itself. The governing principle is: stop and ask, never guess. Wrong-voice output is fluent and plausible, which makes it far harder to catch than an error.

The plugin's references/ folder is no longer scanned. Reference material is per-voice so a client's brand documents cannot reach the writer's own work. references/DROP-MD-FILES-HERE.md is kept only as a redirect note for upgraders.

Two onboarding flows. onboarding/claude-code-mode.md builds the user's own voice, freewrite included. onboarding/subject-mode.md builds someone else's from samples alone — no freewrite is obtainable, confirmation is second-hand, and §0.2 governs every field. Subject profiles record Confirmed by: ghostwriter, not subject.

Per §0.1, a voice profile outranks the style rules in anti-ai-rules.md. If a confirmed voice uses a pattern the rules discourage, the voice wins. Stored samples/ are the exception — they are provenance, not a live override, because §0.1's precedence belongs to a sample pasted in the current conversation.

Versioning

Three files carry a version number and must stay in lockstep:

  • package.json
  • .claude-plugin/plugin.json
  • .claude-plugin/marketplace.json

npm version bumps only the first. Update the other two in the same commit. npm run check fails on drift and runs automatically on npm publish.

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.