بنقرة واحدة
slipway-review
Inspect intent and quality once and report findings without modifying code or creating a repair loop.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
القائمة
Inspect intent and quality once and report findings without modifying code or creating a repair loop.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Resolve genuine human decisions one question at a time after investigating facts.
Explicitly split an Objective into self-contained vertical Changes with native relationships.
Apply authorized code changes and report implementation-time technical activities honestly.
Explicitly materialize clarified requirements as one self-contained Change or one planning Objective.
Explicitly start and drive Slipway's interruptible soft autopilot one structured Action at a time.
| name | slipway-review |
| description | Inspect intent and quality once and report findings without modifying code or creating a repair loop. |
| disable-model-invocation | true |
Use this capability only for a current Review Action or when the user explicitly invokes it directly. It is always read-only.
For a Run Action, compare the current workspace with the immutable Run-start HEAD and fingerprint. List pre-existing dirty and untracked paths separately. A start-to-current difference is only an observation: concurrent user edits, other Runs, or tools may have contributed, so never claim that the current Run or Implement caused it merely because it appeared later.
Inspect two dimensions:
Report concrete findings with locations, details, and uncertainties. The strict Run Outcome must carry action_kind: "review" matching the current Action. Submit only these combinations:
completed with review.result no_findings_reported or findings_reported;partial with review.result inconclusive;error with review.result error.Set pause and implementation to JSON null, include every common Outcome field, and leave suggested_actions empty. not_run is reserved for a CLI review-skip projection.
Do not modify files, return needs_input, ask the user to choose a repair, suggest Implement, issue pass/fail or approval language, pause the Run, or create an automatic repair or re-review loop. Findings always flow to Summary; the user may later choose a separate Change or a new Run.
When invoked directly outside a Run, use the baseline named by the user. If none is named, inspect the current workspace against HEAD and state that it is a standalone baseline, not a Run-start snapshot. Return the same factual review without inventing Run attribution or protocol IDs.