| name | critique |
| description | Use this skill whenever someone wants feedback on a piece of writing without committing to changing it: essays, emails, board policies, internal docs, marketing copy, blog posts, or any prose draft. Trigger on phrases like "critique this", "how does this read?", "what's wrong with this", "give me notes on this", "review this draft", "does this land", "get me a second opinion on this", or "/critique [file]". This is a read-only pass — it never edits the file. If the user wants the writing actually fixed, not just assessed, use `tighten` instead. |
Critique a document
Runs a document through Deirdre (the editor sub-agent, a critic built on Deirdre
McCloskey's Economical Writing) and reports what she finds. This skill only reads and
reports — it never touches the file. For the version that applies the fixes, use tighten.
How it works
See ../../references/deirdre-method.md for the full
method: how the editor sub-agent works, chunking rules, the critique prompts, and the
surgical/substantive distinction.
The workflow
- Read & chunk. Read each document, measure tokens with
tks, and split it into Pass 1
chunks per the chunking rules in the shared method file.
- Pass 1 — line by line. Launch one
editor agent per chunk (parallel batches) with the
granular prompt. Collect the line-by-line notes.
- Pass 2 — whole doc. Launch the holistic pass with the whole-doc prompt.
- Merge & report. Combine Pass 1 + Pass 2 findings, dedupe overlaps, and report per below.
Make no edits, propose none, ask nothing.
Reporting
- Open with one line. "N line notes, M structural. Biggest: ."
- Group by severity, not by location. Lead with the substantive/structural notes (they're
the ones worth acting on), then the word- and sentence-level notes as a shorter list
underneath — quote, verdict, fix. This is feedback to read and decide on, not a diff to
approve, so there's no need for before/after pairs on every line.
- Distinguish the two structural-cut kinds. Semantic duplicates (heading-echoes,
restatements, closing flourishes that only restate) are safe cuts; concreteness-for-brevity
cuts depend on the document's purpose — report those as conditional recommendations. Don't
resolve the purpose question or ask it; that's
tighten's job.
- Name the single change that matters most, drawn from the Pass 2 pass.
- If most of the notes are minor word-level fixes and the user seems to want them applied, say
so: "Most of this is safe tightening — run
tighten and I'll just apply it."
Guardrails
Read-only. Never edits, sends, posts, or touches anything outside its own report.