القائمة
Use after implementation, before shipping, or when the user asks for review. Focus on correctness, regressions, verification gaps, failure handling, security, and performance. Skip style-only noise.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
Use when a V3 or V4 task needs design before implementation. Produce only enough structure to reduce thrash: decisions, boundaries, sequence, and verification.
Use when the user wants the agent to keep working without waiting for input. Autopilot runs continuously: finish the given plan first, then keep making one meaningful verified improvement per iteration until interrupted.
Use after meaningful work to capture confirmed repo-specific knowledge in `.hypersonic/learned.md` and `.hypersonic/history.jsonl`: conventions, gotchas, test patterns, architecture notes, and user preferences.
Classify coding work into Hypersonic V1-V4 tiers, choose the lightest safe workflow, set verification depth, and hand off to focused skills. Use at the start of software tasks or when scope changes.
Use when the user wants to explore, prototype, vibecode, or move fast without heavy process. Build the first useful version quickly, show something real, and iterate from feedback.
Use when work is ready to be committed, pushed, merged, or turned into a PR. Handle final verification, git state, commit shape, branch choice, PR text, and safety checks.
استنادا إلى تصنيف SOC المهني
| name | code-review |
| description | Use after implementation, before shipping, or when the user asks for review. Focus on correctness, regressions, verification gaps, failure handling, security, and performance. Skip style-only noise. |
| parameters | {"velocity":"high","rigor":"medium","max_questions":1,"test_mode":"auto","auto_commit":false} |
Good review catches issues that hurt users, operators, reviewers, or the next engineer. Taste is secondary.
velocity: timebox and lead with the riskiest areas.rigor: include rollout, security, performance, and regression exposure.test_mode: enforce that verification scope before approval.auto_commit: small review fixes may be committed only when the workflow permits it.Check in this order:
Do not start with naming or formatting unless the name hides real behavior.
Check external boundaries:
Ask what happens when the dependency fails, returns bad data, or is slow.
Require fresh evidence before approval:
none: direct behavior check plus risk-driven checks.auto: relevant checks for the changed area.relevant: targeted tests plus nearby regression coverage.full: full suite before final approval.If tests are missing for risky behavior, that is a finding.
If the user asks for QA, match depth to test_mode:
auto: critical and high-risk paths only.relevant: critical, high, and medium-risk paths.full: include polish, accessibility, and edge flows.End with ship-readiness: ready, ready with caveats, or not ready.
Look for confusing control flow, magic values, hidden coupling, and comments missing where the code would otherwise be hard to follow.
Scan for:
Prefer fixing concrete issues when the fix is small and obvious.
When reporting, use:
- [severity] [problem]. [why it matters]. [smallest good next action].
Lead with findings. If there are no findings, say so and mention any verification gap.
When acting on human, reviewer, or agent feedback:
The standard: find real risk, tighten the work, and keep shipping moving.