Skip to content

ValentinNikolaev/agent-skills / agent-plugins/skills/review-arch

review-arch

Verified

Review system boundaries and architecture tradeoffs; not line-level code review.

references/review-cost-examples.md

View on GitHub
# Architecture review cost examples

| Request or condition | Before | After |
|---|---|---|
| Same design appears in a spec, implementation plan, and architecture decision | All reviewers repeat requirement coverage and deadline arithmetic | `review-arch` owns responsibilities, dependency direction, data/lifecycle ownership, failure isolation, and structural alternatives; specification testability belongs to `review-spec`, ordered calculations to `plan-review` |
| Two workers hold pool slots and need additional slots before release | Repeat the plan review's numeric schedule | Identify the circular resource ownership and lack of a release path as a structural critical risk; return `CONCERNS` without a second step-by-step arithmetic derivation |
| Supplied plan analysis shows retries cannot fit the task deadline | Re-derive the entire calculation to support an architecture finding | Reuse the cited result to constrain the affected decision; missing capacity evidence stays explicitly unverified rather than receiving `SOUND` |
| `shared_context=_coord/PROJ-123/code-facts.md` supplies current component paths, excerpts, and table ownership | Rebuild the complete repository model from source | Reuse the boundary facts; inspect only a doubtful ownership claim or missing material evidence and read applicable repository instructions |
| An excerpt predates an ownership transfer or a live configuration change | Treat a supplied fact as authoritative indefinitely | Inspect the disputed boundary or request the fresh external snapshot, retaining source time/revision limits |
| `light`: one-repository UI component-boundary proposal, no concurrency/storage changes, seven concerns including one critical risk | Re-read the proposal's research and emit seven findings | Reuse research, inspect only the affected boundary and doubts, show five most severe findings and two omitted, keep `CONCERNS` and existing scenario/impact/correction fields |
| `light` requested for a durable-state ownership migration | Use light because the proposal is short | Explain ineligibility and use the normal architecture review; do not automatically start a separate specialist review |
| Caller selects a model for a UI/text/docs boundary adjustment versus a concurrency/storage decision | Use the top model for both | Recommend a cheaper model for UI/text/docs changes and the top model for concurrency/storage decisions; model choice remains with the caller |

Routing examples: a question about component ownership and alternatives activates `review-arch`, including with `shared_context=...` or `light`. An ordered implementation plan activates `plan-review`; requirement coverage and acceptance criteria activate `review-spec`; changed-line defects activate `code-review`. Mode and evidence reuse preserve the existing `SOUND`, `CONCERNS`, and `BLOCKED` meanings.
receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementationStars: 323kUpdated Claude CodeCodexVerifiedInstall
requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirementsStars: 323kUpdated Claude CodeCodexVerifiedInstall
using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallbackStars: 323kUpdated Claude CodeCodexVerifiedInstall
verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions alwaysStars: 323kUpdated Claude CodeCodexVerifiedInstall