reviewcrew git:20261014.5b60e38A

Immutable. This exact content never changes and is served at /api/v1/blob/e307a174c055d1af.

---
name: reviewcrew
disable-model-invocation: true
description: 'Run a full codebase audit: spawns all specialist reviewers in parallel and writes a consolidated report to REVIEW_REPORT.md'
allowed-tools: [Agent, Read, Write, Glob, Grep, mcp__plugin_albino_agent-memory__project_search, mcp__plugin_albino_agent-memory__user_search]
---

# Review Crew

Spawn all specialist review agents in parallel, collect their findings, and write a single consolidated report to `REVIEW_REPORT.md` in the project root.

## Step 1: Load Memory

Before spawning any reviewer, call project_search and user_search with relevant terms to load project conventions and user preferences. Use findings to inform how you brief each reviewer and to flag known gotchas or decisions that are relevant to the review.

## Step 2: Spawn All Reviewers in Parallel

Spawn all of the following agents simultaneously. Do not wait for one to finish before starting the next. Each agent prompt MUST begin with:

```
MANDATORY: Read AGENTS.md and follow its rules before doing anything.
```

Agents to spawn in parallel:

1. **security-reviewer**: security vulnerabilities across all categories
2. **code-reviewer**: correctness, style, patterns, anti-patterns
3. **architecture-reviewer**: structure, coupling, separation of concerns, SOLID
4. **performance-reviewer**: bottlenecks, complexity, memory, queries
5. **test-reviewer**: coverage, quality, missing cases, flakiness
6. **logging-reviewer**: logging gaps, monitoring, audit trail
7. **dependency-reviewer**: outdated, vulnerable, unused packages
8. **docs-reviewer**: accuracy, completeness, staleness
9. **agents-md-reviewer**: AGENTS.md rule inconsistencies
10. **accessibility-reviewer**: WCAG compliance, ARIA, keyboard navigation, screen reader compatibility
11. **api-design-reviewer**: REST/GraphQL naming, HTTP semantics, versioning, error shape, backward compatibility
12. **database-reviewer**: schema design, migration safety, indexing, constraints, query patterns
13. **i18n-reviewer**: hardcoded strings, date/number formatting, pluralization, RTL, locale handling
14. **unused-code-reviewer**: unused code, tests, CI, infrastructure, containers, scripts, schema, config, docs, and assets

## Step 3: Collect All Results

Wait for all agents to complete. Collect every finding from every agent. Do not discard or summarize findings: preserve the full output of each reviewer.

## Step 4: Write Consolidated Report

Write the full consolidated report to `REVIEW_REPORT.md` in the project root, structured as follows:

1. A `# Codebase Review Report` title with a `Generated: <current date and time>` line.
2. A `## Summary` section containing a severity table with columns `Review Area | Critical | High | Medium | Low`: one row per reviewer in the Step 2 order, then a bold **Total** row. Add this note above the table: "Each row reflects the count from that reviewer's own output. Cross-reviewer duplicates are not merged."
3. One section per reviewer, in the Step 2 order, separated by `---` rules:

```markdown
## <Review Area> Review

<full output from that reviewer>
```

Use the reviewer's area name for the heading (e.g. `## Security Review`, `## API Design Review`, `## Unused Code Review`, `## AGENTS.md Consistency Review`).

## Rules

- Do not truncate or summarize any reviewer's output: paste full findings
- Fill in the summary table with actual counts from each reviewer's output. Count each issue once per reviewer that reported it: do not merge cross-reviewer duplicates. Write `0` for severity levels where a reviewer found nothing; do not leave cells blank.
- If a reviewer finds no issues, write `No issues found.` under its section: do not omit the section
- If a reviewer fails or errors, write `Reviewer failed: <error>` under its section
- Report file must be complete and self-contained: no references to external files needed to understand findings