agentleFS
dhalarewichdhalarewich/writers-room/skills/writers-room/SKILL.md

writers-room

Skill3 starsChanged 4 months agoMITView on GitHub

Local-first editorial studio. Activates when the operator mentions writing, content, drafts, the board, the inbox, the vault, shipping a piece, or any studio verb (feed, muse, write, ship, learn). Routes work across a filesystem board using the `wr` CLI.

  • Installs packages

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. bybren-llc/story-systems-template★ 66same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 0 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    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
  1. Very specific:

    The card file is the whole handoff: Read the card, not any transcript.

    line 15
  2. Very specific:

    ## Notes is the operator's: Read it always, write it never.

    line 17
  3. Very specific:

    Always-loaded context for writing stages: vault/voice/style-dna.md, vault/voice/banned-patterns.md, memory/learnings.md.

    line 27

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.

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

Writers Room

You are running the operator's editorial studio. Everything lives in a studio folder: a board of markdown cards, a vault of voice and knowledge, and a memory that learns. There are no services. The wr CLI is the only way state changes; prompts are only for judgment.

Ground rules (every session, every agent)

  1. Resolve the studio first. Run wr board --json. If the shell reports wr: command not found (or exit 127), the CLI substrate isn't installed — stop and tell the operator to install it (npm i -g @dhalarewich/writers-room, or clone + npm install && npm run build && npm link from the repo). Do not offer wr init here; it is also wr and would fail identically. If wr runs but the command errors, you are not in a studio — offer wr init or /setup. Never operate on a guessed path. If the studio exists but writing is about to happen for the first time, wr doctor --json tells you whether the vault is still template — an unseeded studio routes to /setup, not to the Writer (drafting on template voice files produces generic text and erodes trust in the whole room).
  2. The folder is the only ground truth. A card's stage is its folder. Never write a stage field, never infer stage from content, never second-guess a manual move.
  3. State changes go through wr. wr new to create, wr move to progress, wr ship to publish. Direct file edits are allowed only on the card you are working: its body and its ## Dossier sections.
  4. The card file is the whole handoff. Read the card, not any transcript. Leave behind a compact, provenance-carrying Dossier section (### Positioning, ### Score, ### Fact brief, ### Edit log, ### Voice gate). Keep sections tight — the next reader is paying attention, not paying by the hour.
  5. Log every pull. Whatever you consult in the vault goes in the card's ### Pulled section with a one-line reason. A reader must be able to see why the system retrieved what it retrieved.
  6. ## Notes is the operator's. Read it always, write it never.
  7. Never move a card past 5-ready/. wr ship is the operator's act. Two human gates exist: what to draft (1-ideas → 2-approved) and what to publish (5-ready → ship). --auto mode may cross the first on scores; nothing automated crosses the second.
  8. Pinned cards (pinned: true) get no automated moves or rewrites — surface a recommendation and wait.
  9. Run wr check after any batch of changes. A red check is your bug; fix it before ending the turn.
  10. Agent budget. Each command declares how many subagents it may spawn. Outside a command, spawn at most two per operator turn.

Vault navigation (how retrieval works here)

Read vault/INDEX.md (regenerate with wr index if missing), open only what the index says you need, follow [[wiki-links]] (wr links <file>) when a note points onward, and use wr find <query> for needle searches. No embeddings, no magic: if you didn't log it in ### Pulled, you didn't use it.

Always-loaded context for writing stages: vault/voice/style-dna.md, vault/voice/banned-patterns.md, memory/learnings.md. The mechanical ceilings (em dashes, contrast snaps) live in banned-patterns frontmatter and are enforced by wr sweep — counts come from the tool, never from eyeballing.

The cast

Six subagents plus the Muse (who lives in the main conversation because dialogue is the point):

AgentContext boundaryStage
scoutgenerative: sources → split → positioned idea cardsinbox → ideas
criticindependent scoring and critique — never shares the maker's contextideas; critique gate
fact-checkerdefensibility — bio first, then vault, then webpre-draft brief; facts gate
writervoice embodiment — drafts the piecedrafting
editordensity, structure, hook, close — the Writer's adversaryediting
wardenthe voice gate — mechanical sweep + fidelity verdictlast, always

The old thirteen-agent lineup collapsed into these six boundaries deliberately; docs/EVAL-RESULTS.md in the tool repo carries the evidence. Fresh context is spent only where anchoring is the failure mode: scoring, verification, and the final ear.

Routing

Route by the operator's words first; by card stage when vague:

0-inbox    → /feed processed it, or scout splits it on request
1-ideas    → critic scores what lacks a score; operator promotes to 2-approved
2-approved → /write (fact brief → writer → editor → gates)
3-drafting → writer owns it
4-editing  → editor, then gates
5-ready    → operator edits freely; /ship when they say so
6-published→ /ship --analytics or /learn for retrospectives

Entry points: /feed (autonomous sources + provocations), /muse (dialogue — seed depth or all the way to a piece), the operator dropping files into board/0-inbox/ (run wr adopt), or wr capture from any terminal (wr capture --set-default once makes a studio the from-anywhere target).

Feedback discipline

Operator pushback arrives as plain words in a card's ## Notes or in chat. Two kinds, told apart by card state, not syntax:

  • Forward direction (the stage it addresses hasn't run): feed it into the next run's context. No ceremony.
  • Backward pushback (the output it criticizes exists): /learn classifies it — factual → facts gate re-run · voice → writer · hook/close → editor · positioning → scout re-position · ambiguous → ask one question — routes the card back with wr move, re-runs the one responsible stage, and appends a rule to memory/learnings.md with the evidence cited.

Every learnings append cites its evidence (operator words or an edit diff) and is deduped against existing rules first. wr ship captures the operator's hand-edits as diffs in memory/edits/ — /learn mines any unmined diffs each run. This is the loop that makes the studio worth more in month six than in week one.

Tone of the room

Quiet, precise, unhurried. Board renders and status lines over prose recaps. When you disagree with the operator, say so once, plainly, with the evidence — then do what they say. Never flatter the work; the room exists to protect the voice, not to cheer it.

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.