/review-perf

name: review-perf
user-invocable: false
version: 4.10.0
author: "Gabriel Trabanco <1969593+gtrabanco@users.noreply.github.com>"
license: MIT
description: >
  Internal performance review pass of the agentic-workflow review pack — composed
  in-turn by review-change and product-audit; not a menu entry. Checks the
  changed paths for algorithmic and resource regressions: N+1s, hot-path
  allocations, asset weight, and leaks. Findings only; never edits code.

Review Performance (internal)

Composed by `review-change` / `product-audit` within their conversation — on any agent, follow this file inline as the routed step. **Findings only; never edits, never refactors.**

Scope

The caller's diff or path/glob — the shape is the finder contract's (`skills/review-implementation/references/FINDER_CONTRACT.md`, this section's owner). State the scope at the top of the returned table.

Checklist (evaluate EVERY item — none is optional; n/a must be stated)

• ✓ No N+1 query/IO pattern on changed data-access paths (loop bodies issuing

queries/requests — cite)

• ✓ Algorithmic complexity of new code paths stated where input can grow (an

unbounded O(n²) on user data is a `WARNING` finding)

• ✓ No blocking IO/sync work added on hot/request paths where the platform is

async

• ✓ Resources opened by the change are closed/released on ALL paths including

errors (files, connections, subscriptions, timers)

• ✓ No unbounded growth introduced (caches without eviction, arrays that only

append, listeners never removed)

• ✓ Web only: bundle/asset impact of the change stated (new dep size, image

weight, lazy-loading where the project uses it) — n/a otherwise

• ✓ Repeated computation hoisted where a loop recomputes an invariant (cite)
• ✓ Pagination/limits on any new listing that reads user-scaled data
• ✓ No premature optimization either: complexity added for speed without a

cited measurement is a finding (the repo forbids overengineering)

• ✓ **Measured evidence when declared** — the project's agent guide declares a

`Performance commands` block with a `bench` command AND the diff touches paths its benchmarks cover: RUN the benchmark on the base branch and on the change, and cite both numbers in Evidence as `<cmd> → base <x> / change <y> (<±z%>)`. A regression beyond the noise band (the project's declared band, else ±5%) is a `WARNING` finding; a delta inside the band is no finding. The declared command failing (non-zero exit) is itself a finding (the gate can't measure) — never silently skipped.

• ✓ **No declared perf commands** → state exactly

`n/a — no declared perf commands` for the item above (never skip it silently), and if the diff adds algorithmic code on input that can grow, add a `SUGGESTION` recommending the project adopt the tooling via `init-workspace`'s Performance tooling round.

Materiality bar

The bar is owned by `skills/review-implementation/references/FINDER_CONTRACT.md` — apply it as written: cite the rule it violates beside the evidence; never pad the table.

Return exactly

`Sev` and `Impact` are the two published closed sets (`BLOCKER | CRITICAL | WARNING | SUGGESTION` and `blocking | docs-only | report-only`); their meanings are owned by `review-implementation` (`CLASSIFY.md` §Severity ladder) — assign a token, never a prose grade.

REVIEW PERF — scope: <scope>

| # | Finding | Sev | Impact | Evidence | Suggested fix |
|---|---------|-----|--------|----------|---------------|
| 1 | <what>  | BLOCKER|CRITICAL|WARNING|SUGGESTION | blocking|docs-only|report-only | <file:line — or, for measured findings, `<cmd> → base <x> / change <y> (<±z%>)`> | <smallest action> |

Checklist: <n> evaluated, <n> pass, <n> findings, <n> n/a (<which + why>)
Summary: <1-2 sentences>
Decision: PASS | FAIL

FAIL if any `BLOCKER`, `CRITICAL` or `WARNING` finding is open; PASS otherwise. `SUGGESTION` findings never block — they route to the caller's triage step.

Done when

• Every checklist item was evaluated with evidence (file:line or command output)

or explicitly marked n/a with the reason.

• The fixed-format block above is returned — nothing more, nothing less — and

no code was changed.