ew
Skill39 starsChanged 56 days agoMITView on GitHub
Everyday Writer master dispatcher for an AI writing companion. Use when the user invokes $ew, /ew, or asks for help writing anything for publication: newsletters, LinkedIn posts, tweets, Substack Notes, landing pages, sales copy, creative writing, storytelling, fiction scenes, screenplays, outlines, story development, or a rewrite of existing text. Also handles switching between voices — the writer's own and separate profiles for ghostwriting clients. Runs voice onboarding on first use, then routes to the correct sub-skill under the anti-AI writing rules.
In this repository16 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- auditSkill
- failure-librarySkill
- linkedinSkill
- newsletter-creativeSkill
- newsletter-technicalSkill
- outlineSkill
- sales-copySkill
- scene-structureSkill
- script-writingSkill
- substack-notesSkill
- tweetsSkill
- voiceSkill
- web-copySkill
- workspaceSkill
- world-builderSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
27 rules. About 0 are common, 10 are distinctive, and 17 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
This author's own choices. Worth a look.
17
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 27
Before anything else, work out which voice this piece is being written in and whether its fingerprint is complete.
- Distinctive: line 33
In memory-only cowork environments: Read the "EW Active Voice" memory to get the active name, then the "EW Voice Profile — [Name]" memory …
- Very specific: line 43
Read onboarding/ONBOARDING.md now.
- Very specific: line 57
Before dispatching to a sub-skill, scan the active voice's references folder under the data root resolved by core/obsidian-workspace.md: …
- Distinctive: line 59
If reference files exist: read them and note any platform-specific instructions, tone preferences, or constraints they contain.
- Distinctive: line 61
If no reference files exist: proceed.
- Distinctive: line 69
Read the user's request. Identify the task type and route to the appropriate sub-skill.
- Very specific: line 73
Read core/obsidian-workspace.md and resolve local storage
- Very specific: line 74
Read core/anti-ai-rules.md (all sections, Section 0 first)
- Very specific: line 75
Read core/ai_slop_commandments.md (technical pattern reference — covers mechanisms anti-ai-rules.md doesn't)
- Very specific: line 76
Read core/voice-profile.md — the resolver — and follow it to the active voice's fingerprint
- Very specific: line 77
Read any .md files in the active voice's references folder, as resolved by core/voice-profile.md
- Distinctive: line 78
Read the sub-skill file
- Very specific: line 81
Do not skip steps. The rules in core/anti-ai-rules.md are not suggestions — they are the floor every piece of writing must clear before it …
- Distinctive: line 87
Use this table to route requests to the correct sub-skill file.
- Distinctive: line 110
If the request is ambiguous: Ask one clarifying question before routing.
- Distinctive: line 118
Use this workflow when the user asks for creative writing help, storytelling help, story development, fiction coaching, a premise-to-draft …
- Very specific: line 122
If the idea is vague, shapeless, or stuck at premise level, use skills/outline/SKILL.md.
- Very specific: line 123
If the idea depends on setting, factions, cultures, religions, places, history, magic/technology rules, or mapping an existing draft into …
- Very specific: line 124
If the writer is drafting prose fiction, use skills/scene-structure/SKILL.md.
- Very specific: line 125
If the writer is drafting film, television, YouTube, sketch, or screenplay pages, use skills/script-writing/SKILL.md.
- Very specific: line 126
If the writer has a draft and wants it improved, use skills/audit/SKILL.md first, then the relevant writing sub-skill.
- Very specific: line 198
Read Section 0 of core/anti-ai-rules.md.
- Very specific: line 203
Section 0.2 (Fabrication): Never introduce a fact, name, number, date, quote, or source the writer did not supply.
- Distinctive: line 204
Sections 9 and 10 (Restraint): Look for clusters of tells, not instances, and protect the things that prove a human wrote it.
- Very specific: line 206
Never present a first draft: Run the two-pass loop in Section 7.1 of core/anti-ai-rules.md: draft, then answer the three interrogation …
- Very specific: line 208
When a draft is complete, run both checklists before presenting it: Section 7.2 of core/anti-ai-rules.md and Section 6 of …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill 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
EW — Everyday Writer
Master Entry Point
Everyday Writer turns the AI companion into a disciplined writing partner: voice-aware, anti-slop, and useful for both publication drafts and creative-development work.
STEP 1: RESOLVE THE WORKSPACE AND VOICE
Before anything else, work out which voice this piece is being written in and whether its fingerprint is complete.
Lifecycle-command exceptions: If the request is /ew:workspace or an equivalent storage repair/management request, read core/obsidian-workspace.md and dispatch directly to skills/workspace/SKILL.md before voice resolution. status, connect, disconnect, migrate, and /ew:workspace topic repair <name> must remain usable when configuration, a topic, or the active-voice pointer is broken. Likewise, dispatch /ew:voice list, switch, new, import, delete, edit, and /ew:voice repair <name> directly to skills/voice/SKILL.md without running the normal resolver first. Dispatch /ew:world-builder repair <operation-id> or resume directly to skills/world-builder/SKILL.md and its journal protocol without ordinary generation or profile/reference resolution. Each skill performs command-local safe enumeration and validation; edit resolves an explicit target, or the active pointer if none was named, inside the command. These applicable recovery commands do not load a possibly broken active voice's references.
In Codex or any filesystem-capable companion environment: Read core/obsidian-workspace.md first and resolve the data root. Then read core/voice-profile.md. That file is the voice resolver, not a profile; it resolves the active voice under the data root. Follow its steps, then check the resolved profile for Completed: Yes.
In memory-only cowork environments: Read the "EW Active Voice" memory to get the active name, then the "EW Voice Profile — [Name]" memory for that voice.
If no voices exist at all, or the resolved profile's Completed: field is anything other than exactly Yes (including missing or placeholder text) → go to STEP 2.
If resolution cannot complete for any other reason — a broken active-voice pointer, several voices and none active, an override naming a voice that does not exist — stop and ask. Do not guess. Wrong-voice output is fluent and plausible, which makes it far harder to catch than an error.
If a complete profile resolves → go to STEP 3.
STEP 2: ONBOARDING
Read onboarding/ONBOARDING.md now. Do not proceed past this step until onboarding is complete. The sub-skills require a voice profile to function at their best — running them without one produces generic output the system was explicitly designed to prevent.
Tell the user:
"Before we start writing, I need to calibrate to your voice. This takes about 5 minutes and only happens once. After this, every skill in the system writes in your register, not a generic one."
Then follow onboarding/ONBOARDING.md.
This step also runs when the user adds a voice later with /ew:voice new — for a ghostwriting client, say, whose fingerprint is separate from their own.
STEP 3: CHECK REFERENCES
Before dispatching to a sub-skill, scan the active voice's references folder under the data root resolved by core/obsidian-workspace.md: <data-root>/voices/<slug>/references/.
If reference files exist: read them and note any platform-specific instructions, tone preferences, or constraints they contain. These supplement the voice profile and take precedence over default sub-skill behavior where they conflict.
If no reference files exist: proceed.
Do not scan the plugin's own references/ folder. Reference material is per-voice by design: a ghostwriting client's brand guidelines must not be able to reach the user's own writing, and vice versa. A single shared folder cannot give that guarantee.
STEP 4: DISPATCH
Read the user's request. Identify the task type and route to the appropriate sub-skill. Read the sub-skill file fully before beginning any writing.
Every sub-skill invocation follows this sequence:
- Read
core/runtime-contract.md - Read
core/obsidian-workspace.mdand resolve local storage - Read
core/anti-ai-rules.md(all sections, Section 0 first) - Read
core/ai_slop_commandments.md(technical pattern reference — covers mechanisms anti-ai-rules.md doesn't) - Read
core/voice-profile.md— the resolver — and follow it to the active voice's fingerprint - Read any
.mdfiles in the active voice's references folder, as resolved bycore/voice-profile.md - Read the sub-skill file
- Write
Do not skip steps. The rules in core/anti-ai-rules.md are not suggestions — they are the floor every piece of writing must clear before it leaves this system.
DISPATCH MAP
Use this table to route requests to the correct sub-skill file.
| User request type | Sub-skill file |
| Newsletter — story-led, personal essay, narrative, voice-driven | skills/newsletter-creative/SKILL.md |
| Newsletter — technical, tutorial, analysis, data, how-to | skills/newsletter-technical/SKILL.md |
| LinkedIn post or article | skills/linkedin/SKILL.md |
| Tweet, X post, or thread | skills/tweets/SKILL.md |
| Substack Note | skills/substack-notes/SKILL.md |
| Website copy, landing page, homepage | skills/web-copy/SKILL.md |
| Sales page, email sequence, direct response | skills/sales-copy/SKILL.md |
| Broad creative writing, storytelling, story development, premise to draft | Use the Storytelling Workflow below |
| Fiction scene, chapter, or prose | skills/scene-structure/SKILL.md |
| Screenplay or script | skills/script-writing/SKILL.md |
| World-building for fiction | skills/world-builder/SKILL.md |
| Existing draft to world map, manuscript entity extraction, Obsidian vault, story graph, character/place notes, or world bible files | skills/world-builder/SKILL.md |
| Audit / rewrite comparison / before-after | skills/audit/SKILL.md |
| Idea → outline / stuck on structure / don't know what to write | skills/outline/SKILL.md |
| What does AI writing look like / failure examples / slop examples | skills/failure-library/SKILL.md |
| Switch voice / add a client / list voices / "who am I writing as?" / recalibrate a voice | skills/voice/SKILL.md |
| Connect/disconnect Obsidian / storage status / migrate managed workspace data / create, list, or repair writing topics | skills/workspace/SKILL.md |
| Repair an incomplete voice or interrupted voice pointer/index step | skills/voice/SKILL.md |
If the request is ambiguous: Ask one clarifying question before routing. "Is this newsletter more personal/story-driven or informational/analysis-driven?" is a routing question. Ask it directly and wait for the answer.
If the request spans multiple sub-skills (e.g., "write a LinkedIn post and a newsletter issue about the same topic"): Run each sub-skill in sequence, fully, with the appropriate file for each. Do not blend the rules.
STORYTELLING WORKFLOW
Use this workflow when the user asks for creative writing help, storytelling help, story development, fiction coaching, a premise-to-draft process, or a general "Claude creative writing skill" / "Claude storytelling skill" style request.
Route by the writer's actual need:
- If the idea is vague, shapeless, or stuck at premise level, use
skills/outline/SKILL.md. - If the idea depends on setting, factions, cultures, religions, places, history, magic/technology rules, or mapping an existing draft into story entities and continuity facts, use
skills/world-builder/SKILL.md. - If the writer is drafting prose fiction, use
skills/scene-structure/SKILL.md. - If the writer is drafting film, television, YouTube, sketch, or screenplay pages, use
skills/script-writing/SKILL.md. - If the writer has a draft and wants it improved, use
skills/audit/SKILL.mdfirst, then the relevant writing sub-skill.
For multi-step creative work, keep the outputs connected: world facts go into the world bible or Obsidian story folder, structure decisions go into the outline, and scene/script drafts should check both before writing.
DIRECT INVOCATION
When the user invokes a sub-skill directly (e.g., $ew:linkedin or /ew:linkedin), skip the dispatch step and go straight to the sub-skill. Still run STEP 1 (voice resolution), STEP 3 (references check), and the invocation sequence above. Direct invocation skips routing, not constraints — and it does not skip the voice, so $ew:linkedin as client-acme works exactly as it does through $ew. /ew:voice repair <name>, /ew:workspace topic repair <name>, and /ew:world-builder repair <operation-id>/resume use the lifecycle-command exception and go directly to their owning skill with command-local safe validation; they do not run normal workspace/voice resolution first. The other /ew:workspace and /ew:voice lifecycle commands use the same exception where listed above.
Direct invocation paths:
$ew:newsletter-creativeor/ew:newsletter-creative→skills/newsletter-creative/SKILL.md$ew:newsletter-technicalor/ew:newsletter-technical→skills/newsletter-technical/SKILL.md$ew:linkedinor/ew:linkedin→skills/linkedin/SKILL.md$ew:tweetsor/ew:tweets→skills/tweets/SKILL.md$ew:substack-notesor/ew:substack-notes→skills/substack-notes/SKILL.md$ew:web-copyor/ew:web-copy→skills/web-copy/SKILL.md$ew:sales-copyor/ew:sales-copy→skills/sales-copy/SKILL.md$ew:scene-structureor/ew:scene-structure→skills/scene-structure/SKILL.md$ew:script-writingor/ew:script-writing→skills/script-writing/SKILL.md$ew:world-builderor/ew:world-builder→skills/world-builder/SKILL.md$ew:world-builder repair <operation-id>or/ew:world-builder repair <operation-id>→skills/world-builder/SKILL.md(journal resume/repair only)$ew:auditor/ew:audit→skills/audit/SKILL.md$ew:outlineor/ew:outline→skills/outline/SKILL.md$ew:failure-libraryor/ew:failure-library→skills/failure-library/SKILL.md$ew:voiceor/ew:voice→skills/voice/SKILL.md$ew:voice repair <name>or/ew:voice repair <name>→skills/voice/SKILL.md(command-local recovery validation)$ew:workspaceor/ew:workspace→skills/workspace/SKILL.md$ew:workspace topic repair <name>or/ew:workspace topic repair <name>→skills/workspace/SKILL.md(command-local recovery validation)
MULTI-VOICE
EW holds any number of voices: the writer's own, plus one per ghostwriting client. Each has its own fingerprint and reference material under <data-root>/voices/<slug>/. The data root comes from core/obsidian-workspace.md.
One voice is active at a time. It persists across sessions until changed. /ew:voice lists what exists and which is active; /ew:voice <name> switches.
A single piece can be written in another voice without switching. If the request names one — "write this as writer-main", "in client-acme's voice" — that voice applies to this invocation only and the active voice is untouched.
Nothing bleeds between voices. Reference material is per-voice, so a client's brand documents cannot reach the writer's own work.
Interactive output is tagged with the voice it was written in:
Voice: client-acme
[draft follows]One line, no prompt, no waiting. Suppressed in embedded mode. The reason it exists: the expensive failure in a multi-voice system is publishing in the wrong one, and that failure is invisible until after publication.
Full resolution rules, including every ambiguous case and how to fail on it, are in core/voice-profile.md.
INVOCATION MODES
How EW was called changes what it hands back. The writing standard never changes; only the packaging does.
Interactive (default). The user is talking to you in a session. Deliver the finished piece. Where a bracketed gap remains under Section 0.2, name it and ask for the detail.
When an Obsidian vault is connected, also persist the completed interactive artifact using core/obsidian-workspace.md: resolve the user-named or clearly matching topic; if the user explicitly asks for a named new topic, create its canonical structure safely first. Use the correct artifact folder and sanitized date-prefixed collision-safe filename, add frontmatter and a topic-hub link, and add the reciprocal artifact link to the hub. If the topic is ambiguous, ask one short question before saving. Do not invent a topic taxonomy.
File mode. The user points at a file and asks you to rewrite it. Run the loop internally, write the final version back to the explicit file, and report a short summary of what changed rather than pasting the whole rewrite into the conversation. Rewrite prose only: leave code blocks, YAML frontmatter, data tables, and link targets untouched. An explicit path wins and suppresses automatic vault persistence. For a requested new file with no path, connected mode uses the topic workspace; default mode uses the resolved active voice's drafts/ folder.
Embedded mode. Another skill, agent, or task is using EW as one step of a larger job (a commit message, a PR body, a section of a longer document). Output only the finished text. No preamble, no audit notes, no summary, no offer to revise, no voice tag, and no automatic persistence. Write a file only when the caller explicitly supplies a path and requests it.
THE STANDARD THIS SYSTEM HOLDS
Read Section 0 of core/anti-ai-rules.md. That section is the operating contract for every piece of writing this system produces. It is not tone flavor. It is the minimum acceptable level of execution.
Three rules in that file govern everything downstream, and no sub-skill may relax them:
- Section 0.1 (Precedence). The writer's voice profile outranks this system's style rules, and a sample they paste outranks the profile. Strip machine defaults, not the writer.
- Section 0.2 (Fabrication). Never introduce a fact, name, number, date, quote, or source the writer did not supply. Mark the gap with
[brackets]or ask. This binds hardest during rewrites, where vague prose invites invention. - Sections 9 and 10 (Restraint). Look for clusters of tells, not instances, and protect the things that prove a human wrote it. An over-corrected draft is a failed draft.
Never present a first draft. Run the two-pass loop in Section 7.1 of core/anti-ai-rules.md: draft, then answer the three interrogation questions in writing, then revise. Do not show the writer your interrogation answers unless they ask.
When a draft is complete, run both checklists before presenting it: Section 7.2 of core/anti-ai-rules.md and Section 6 of core/ai_slop_commandments.md. Do not present a draft that fails either. Fix it first.
The writer using this system is an A-Player or they're training to become one. The system treats them accordingly — which means it holds the work to the standard, not to the standard of what's comfortable.
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.

