The content below is the complete file Claude Code will execute — copy or download it to install.

# /critique — Contrarian Evaluator Fast, single-pass adversarial critique. Finds fatal flaws before you act. ## Instructions ### Step 0 — Resolve the target Use the first available: 1. `$ARGUMENTS` if present 2. Selected text from the IDE 3. Latest assistant-produced plan, code block, or diff in the session 4. If none found: reply "No target found. Provide text or select content." and stop. ### Step 1 — Adversarial critique Persona: skeptical senior expert whose only job is to find flaws. Do NOT suggest fixes yet. Rules (enforce strictly): - Every finding must **quote or reference the exact claim, step, or line** it attacks — no floating objections - State the **concrete consequence** if the flaw is ignored — not "this could be a problem", but what specifically breaks and when - A flaw is "fatal" only if it **plausibly changes the go/no-go decision** — not "worth fixing someday" - **Limit fatal flaws to 1–3 highest-impact issues** — more than 3 signals genericity, not rigor - Disallow generic boilerplate (e.g. "consider edge cases", "add error handling") not anchored to the artifact Output exactly: <critique> ## Fatal flaws (1–3 max) [Each entry: quoted/referenced claim → concrete consequence if ignored] ## Significant weaknesses [Real issues, not blockers. Quoted reference + consequence.] ## Unstated assumptions [Implicit dependencies or preconditions the artifact relies on without stating] ## Best opposing case [The strongest single argument against the entire approach — not a list of nitpicks] </critique> ### Step 2 — Minimal fix list For each **fatal flaw only**: the smallest change that resolves it. One bullet per flaw. ### Step 3 — Self-check Re-read the critique. Add any meaningful omissions — or explicitly state "No additional omissions found." Do not manufacture findings. ### Step 4 — Verdict State exactly one: - **PROCEED** — zero fatal flaws found - **REVISE** — at least one fatal flaw; each has a plausible bounded fix (per Step 2) - **ABORT** — flaw is structural, or fixes would change the core proposal

/critique — Contrarian Evaluator

Fast, single-pass adversarial critique. Finds fatal flaws before you act.

Instructions

Step 0 — Resolve the target

Use the first available:

$ARGUMENTS if present
Selected text from the IDE
Latest assistant-produced plan, code block, or diff in the session
If none found: reply “No target found. Provide text or select content.” and stop.

Step 1 — Adversarial critique

Persona: skeptical senior expert whose only job is to find flaws. Do NOT suggest fixes yet.

Rules (enforce strictly):

Every finding must quote or reference the exact claim, step, or line it attacks — no floating objections
State the concrete consequence if the flaw is ignored — not “this could be a problem”, but what specifically breaks and when
A flaw is “fatal” only if it plausibly changes the go/no-go decision — not “worth fixing someday”
Limit fatal flaws to 1–3 highest-impact issues — more than 3 signals genericity, not rigor
Disallow generic boilerplate (e.g. “consider edge cases”, “add error handling”) not anchored to the artifact

Output exactly:

## Fatal flaws (1–3 max) [Each entry: quoted/referenced claim → concrete consequence if ignored]

Significant weaknesses

[Real issues, not blockers. Quoted reference + consequence.]

Unstated assumptions

[Implicit dependencies or preconditions the artifact relies on without stating]

Best opposing case

[The strongest single argument against the entire approach — not a list of nitpicks]

Step 2 — Minimal fix list

For each fatal flaw only: the smallest change that resolves it. One bullet per flaw.

Step 3 — Self-check

Re-read the critique. Add any meaningful omissions — or explicitly state “No additional omissions found.” Do not manufacture findings.

Step 4 — Verdict

State exactly one:

PROCEED — zero fatal flaws found
REVISE — at least one fatal flaw; each has a plausible bounded fix (per Step 2)
ABORT — flaw is structural, or fixes would change the core proposal