/critique

/soe:critique — adversarial review of a design or plan

Runs the adversarial gate (design §3.4) on demand. It dispatches the soe:devils-advocate agent (Opus, fresh isolated context) to red-team a target document using the soe:adversarial-review methodology, and returns a numbered findings list followed by the discuss all / some / continue choice.

Usage

/soe:critique design <file>    # red-team a design/spec/architecture doc
/soe:critique plan <file>      # red-team a plan doc + design↔plan cross-reference

• design <file> — design mode: gaps, inconsistencies, missing pieces, pattern misuse, against the quality lens.
• plan <file> — plan mode: everything design mode does on the plan, PLUS a design↔plan cross-reference (faithfulness, drift, dropped scope, scope creep). Point it at the plan; it locates and reads the referenced design.

Examples

/soe:critique design docs/plans/2026-07-03-soe-design.md
/soe:critique plan docs/plans/my-feature-plan.md

What it does

1. Parse the mode (design | plan) and the target <file> from the arguments. If the mode is missing or not one of design/plan, or the file does not exist, report the correct usage and stop.
2. Dispatch soe:devils-advocate via the Task tool, instructing it to run the soe:adversarial-review skill in the parsed mode against <file> (and, for plan mode, to locate and cross-reference the design doc).
3. Relay the agent's numbered findings and its discuss all / discuss some / continue prompt back to you. Do not resolve findings until you choose.

This is the same methodology the orchestrator runs automatically at EVALUATE_PLAN; /soe:critique is the manual, on-demand entry point.