sethshoultes/authors-critique

/authors-critique <file> [author1 author2 ...]

Fast, cheap gut-check critique.

When to use

• You want a directional read on a draft before investing in a full edit pass.
• You're deciding which authors to run /authors-edit with — use critique to triage.
• You want to see which authors have the strongest reaction before committing to a markup pass.

Not for: marked-up passages (that's /authors-edit); craft debates between two authors (that's /authors-debate).

Instructions for Claude

1. Parse arguments: file path (required) + optional author names. Same parsing as /authors-edit.

2. Verify the file exists.

3. If no authors named, pick 3 using the same genre-signal logic as /authors-edit, but with a wider net. Default triad: Hemingway + Orwell + Didion for ambiguous cases (covers sentence-level, argument clarity, observational specificity).

4. Fan out via Agent tool. Dispatch all authors in parallel in a single message. For each:

   • subagent_type: <author>-persona
   • model: "haiku" (override — critique is opinion-style and tolerates the cheaper model; edit and debate stay on Sonnet)
   • Prompt:

     CRITIQUE MODE - TERSE OUTPUT ONLY.
     
     Read this draft and respond with EXACTLY 3 bullets. Each bullet is one sentence. No introduction. No markdown markup of passages. No rewrites. Just the three most important things you notice.
     
     End with one line: "If I'm not the right voice here, try <X>." — or omit if you are.
     
     Draft:
     <full file contents>
     

5. Consolidate:

   # /authors-critique on <filename> — <author A>, <author B>, <author C>
   
   ## <Author A>
   - <bullet 1>
   - <bullet 2>
   - <bullet 3>
   
   ## <Author B>
   - <bullet 1>
   - <bullet 2>
   - <bullet 3>
   
   ## <Author C>
   - <bullet 1>
   - <bullet 2>
   - <bullet 3>
   
   ## Consensus
   (one sentence naming what all or most authors flagged)
   
   ## Sharpest disagreement
   (one sentence naming the most productive disagreement, or "no significant disagreement" if so)
   
   ## Handoffs
   (if any author suggested a different voice, name them here)
   

6. Output to stdout. No manuscript changes.

Notes

• This skill is cheap by design. Resist the temptation to pad the output.
• If any author returns more than 3 bullets, trim their output to 3 in consolidation — report verbatim otherwise.
• Sub-agents inherit cwd; bible files are read automatically via each persona's protocol.
• Model: each sub-agent dispatch includes model: "haiku" as an override. This is intentional — critique is opinion-style work that doesn't require Sonnet-level reasoning. If you find critiques losing quality on Haiku (e.g., personas drift off-voice, cross-references hallucinate), drop the override in the Agent call and the dispatch falls back to the agent's default model: sonnet.