critique
Deliver a harsh, evidence-backed critique of whatever the user points at — a file, directory, codebase, skill, design doc, API, spec, README, or pasted text.
Skill by folkengine
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/folkengine-critique
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: documents, design, backend
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/critique
What's inside
- Skill name
critique- Description
- Deliver a harsh, evidence-backed critique of whatever the user points at — a file, directory, codebase, skill, design doc, API, spec, README, or pasted text. Use when the user types `/critique <target>`, or asks to "tear this apart", "roast this", "be brutal", "harsh critique", "no-holds-barred review", "red-team this doc", "what's wrong with this", "poke holes in this", "devil's advocate this" — even if they never say critique. Do NOT trigger for the standardized scored audits (/muratori, /reusability, /smimd, /untangle, /conformance) or for a correctness-focused diff review (/code-review); this is a free-form adversarial read of one artifact.
- License
- Apache-2.0
- Instructions
- 68 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# /critique
Deliver an adversarial, unhedged critique of one artifact. The reader is the
artifact's author and wants the truth, not encouragement. Harsh means
**unhedged and specific** — attack the work, never the author; no praise
sandwich, no "consider perhaps maybe".
Honesty outranks harshness: every finding must survive your own verification,
and a genuinely strong target gets a short, plainly positive verdict. A forced
hit-piece is a failed critique.
## Method
1. **Read everything the target touches.** The named file plus its scripts,
assets, references, templates, callers, siblings — critique the real
artifact, not a summary of it. For pasted text, the paste is the whole
target; say so in the verdict.
2. **Steelman first.** Before attacking, state (to yourself) the strongest
version of what the artifact is trying to do. Judge it against that, so
the critique cannot be dismissed as missing the point.
3. **Judge against the target's own claims and its peers.** Promises in its
docstring/frontmatter/README are contracts — a claim the artifact breaks
is your best material. If it lives among siblings (a skills folder, a
workspace), their conventions are the house standard it is measured by.
4. **Verify every hit yourself** — open the file, run the command, reproduce
the failure. A finding carries file:line or a direct quote. Can't verify
it? Cut it, or mark it `SUSPECTED` inline.
5. **Deliver in the conversation.** Write a file only when the user asks for
one.
## Report shape
The critique is these four parts, in this order:
1. **Verdict** — one blunt opening paragraph: what this artifact is, whether
it does what it claims, and the single worst problem. The reader who stops
here has the answer.
2. **Charge sheet** — findings ordered worst-first, each tagged with a
severity anchor:
- `FATAL` — defeats the artifact's stated purpose, or fails on first real
use.
- `SERIOUS` — wrong or misleading in a way that will bite a real user.
- `MINOR` — friction, inconsistency, polish.
Each finding: the claim in one sentence, the verified evidence
(file:line / quote / reproduced output), and the consequence — who gets
hurt and when. Severity comes from consequence, not from how easy the
finding was to spot.
3. **What survives** — the parts that genuinely hold up, in two or three
sentences. This is triage information ("don't rewrite these parts"), not
consolation.
4. **Fix order** — a short leverage-ordered list: the fewest changes that
repair the most, starting with the FATALs.
No numeric scores, no letter grades — the verdict paragraph is the headline.
## Common mistakes
| Mistake | Fix |
|---|---|
| Critiquing only the named file | Read its scripts/assets/references/callers first; the rot is usually in the seams |
| Harshness as snark or adjectives | Harshness is severity + evidence; every insult-shaped sentence must be replaceable by a file:line |
| Inventing findings to seem tough | If it holds up, the verdict says so plainly; honesty outranks harshness |
| Unverified claims stated as fact | Verify or mark `SUSPECTED`; one wrong finding discredits the ten right ones |
| Flat list of equal-weight gripes | Anchor every finding FATAL / SERIOUS / MINOR and order worst-first |
| Grading on a curve of intent | Judge against the artifact's own promises and its siblings' conventions |
| Burying the verdict at the end | The verdict is the opening paragraph, not a closing summary |From folkengine/.claude, Apache-2.0.
More from folkengine/.claude
- 0.570score
huxley
folkengine
Use when the user types /huxley <code>, or asks how a function, type, file, or snippet evolved over time, wants its git lineage or "family tree", asks where…
SkillAFreshness: AClaude Code +9 - 0.529score
dev-playbook
folkengine
Use when the user types /dev-playbook, asks to start, scaffold, bootstrap, or init a new project/repo/library in rust, python, go, java, cpp, c, php, ts-lib…
SkillBFreshness: BClaude Code +9 - 0.529score
conformance
folkengine
Measure a codebase against an externally-authored ruleset — a standards-body spec, RFC, protocol document, regulation, house rules PDF, or API contract —…
SkillBFreshness: BClaude Code +9 - 0.529score
untangle
folkengine
Evaluate how entangled a repository is with each of its dependencies and what removing them would take — a standardized audit written to the repo's docs…
SkillBFreshness: BClaude Code +9 - 0.529score
smimd
folkengine
Evaluate a codebase for parallelism opportunities — SIMD lanes, SWAR bit-tricks, MIMD/thread-level data parallelism, and the algorithm or data-structure…
SkillBFreshness: BClaude Code +9 - 0.524score
qanda
folkengine
Generate a paired self-test Q&A study set — <TOPIC>_QUESTIONS.md and <TOPIC>_ANSWERS.md in the repo's docs folder — that helps developers learn how the code…
SkillBFreshness: BClaude Code +9 - 0.524score
quality-commitments
folkengine
Facilitates Quality Commitments work — collaborative definition of quality standards, generation of a Quality Commitments Matrix, evidence-based audit of an…
SkillBFreshness: BClaude Code +9 - 0.519score
debrief
folkengine
Bullet-point summary of what changed on the current branch versus main.
SkillBFreshness: BClaude Code +9 - 0.499score
blooper
folkengine
Use when the user asks for a "blooper report" or wants to dig through a repo's git history to find its most embarrassing early coding mistakes and see which…
SkillBFreshness: BClaude Code +9 - 0.479score
deconstruct
folkengine
Distill an existing codebase into a portable regeneration pack — language-agnostic DECON epics in the /epic house style plus golden test vectors extracted by…
SkillCFreshness: CClaude Code +9 - 0.479score
codebase-kata
folkengine
Turn a real codebase into a TDD-style coding kata that teaches the core technology or algorithm behind it.
SkillCFreshness: CClaude Code +9 - 0.474score
dog
folkengine
Dog mode, aka "walking the dog" — Claude writes no code; it walks the developer through a coding task one step at a time while the developer types everything.
SkillCFreshness: CClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Aug 18, 2026 (2 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- Apache-2.0
- Repository
- github.com/folkengine/.claude
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from folkengine
- 0.570score
huxley
folkengine
Use when the user types /huxley <code>, or asks how a function, type, file, or snippet evolved over time, wants its git lineage or "family tree", asks where…
SkillAFreshness: AClaude Code +9 - 0.529score
conformance
folkengine
Measure a codebase against an externally-authored ruleset — a standards-body spec, RFC, protocol document, regulation, house rules PDF, or API contract —…
SkillBFreshness: BClaude Code +9 - 0.529score
smimd
folkengine
Evaluate a codebase for parallelism opportunities — SIMD lanes, SWAR bit-tricks, MIMD/thread-level data parallelism, and the algorithm or data-structure…
SkillBFreshness: BClaude Code +9 - 0.529score
untangle
folkengine
Evaluate how entangled a repository is with each of its dependencies and what removing them would take — a standardized audit written to the repo's docs…
SkillBFreshness: BClaude Code +9 - 0.529score
dev-playbook
folkengine
Use when the user types /dev-playbook, asks to start, scaffold, bootstrap, or init a new project/repo/library in rust, python, go, java, cpp, c, php, ts-lib…
SkillBFreshness: BClaude Code +9 - 0.524score
qanda
folkengine
Generate a paired self-test Q&A study set — <TOPIC>_QUESTIONS.md and <TOPIC>_ANSWERS.md in the repo's docs folder — that helps developers learn how the code…
SkillBFreshness: BClaude Code +9
Cite this page
critique skill. Agent Mart. https://agentmart.dev/skills/folkengine-critique (last updated 2026-10-24)