{
  "markdown": "# gemini-review\n\nGemini CLI extension for code review with sub-agent delegation. Splits reviews into three specialized dimensions (security, correctness, quality), then synthesizes findings with confidence scoring and progressive disclosure.\n\nMethodology ported from [quorum-counsel](../plugins/quorum-counsel/) and Claude Code's pr-review-toolkit.\n\n## Installation\n\n```bash\n# From the ai-tools repo\ngemini extensions link ./extensions/gemini-review\n\n# Or from a git URL\ngemini extensions install https://github.com/ShravanSunder/gemini-review-agents\n```\n\n### Prerequisites\n\nEnable experimental sub-agents in Gemini CLI settings:\n\n```json\n{\n  \"experimental\": {\n    \"enableAgents\": true\n  }\n}\n```\n\nWithout this setting, the `/review` commands will still work but Gemini will handle the review as a single pass instead of delegating to specialized sub-agents.\n\n## Commands\n\n| Command | Scope | Usage |\n|---------|-------|-------|\n| `/review` | Uncommitted changes (staged + unstaged) | `/review` |\n| `/review:branch` | Current branch vs base branch | `/review:branch main` |\n| `/review:commit` | Specific commit | `/review:commit a1b2c3d` |\n| `/review:pr` | GitHub pull request | `/review:pr 123` |\n\n### Examples\n\n```\n# Review your working directory changes\n/review\n\n# Review everything on this branch vs main\n/review:branch main\n\n# Review a specific commit\n/review:commit HEAD~1\n\n# Review a GitHub PR\n/review:pr 456\n```\n\n## Sub-Agents\n\nEach review delegates to three specialized sub-agents:\n\n| Agent | Focus | Categories |\n|-------|-------|------------|\n| **security-reviewer** | Injection, auth bypass, data exposure, SSRF, silent failures | `security`, `silent-failure` |\n| **correctness-reviewer** | Logic errors, null safety, type issues, edge cases, error handling | `bug`, `error-handling` |\n| **quality-reviewer** | Pattern compliance, test gaps, comment accuracy, performance | `pattern`, `test-gap`, `comment-rot`, `performance` |\n\n## Methodology\n\n### Severity Scale\n\n| Level | Meaning |\n|-------|---------|\n| **P0 / CRITICAL** | Exploitable security, data loss, authz bypass, crash in common path |\n| **P1 / HIGH** | High-likelihood bug, major perf regression, concurrency hazard, broken API contract |\n| **P2 / MEDIUM** | Edge-case bug, maintainability risk, incomplete error handling, moderate test gap |\n| **P3 / LOW** | Minor cleanup, nits only if they materially improve clarity |\n\n### Confidence Scoring\n\nEvery finding is scored 0-100. Only findings with **confidence >= 80** are reported.\n\nWhen two or more sub-agents flag the same issue (same file:line, similar description), a **+15 consensus bonus** is applied (cap at 100). This rescues borderline findings that multiple reviewers independently identified.\n\n### Verdict\n\n| Verdict | Criteria |\n|---------|----------|\n| **PASS** | 0 findings at P0-P1, 2 or fewer at P2 |\n| **PASS WITH CONCERNS** | 0 P0, any P1 or P2 |\n| **REVISE** | Any P0, or 3+ findings at P1 |\n\n### Ignore List\n\nThe following are NOT reported:\n- Pre-existing issues not introduced in this changeset\n- Issues linters/typecheckers/CI would catch\n- Pedantic nitpicks a senior engineer wouldn't flag\n- Issues on unmodified lines\n- Intentional functionality changes\n- Issues silenced by lint-ignore comments\n\n## Output\n\n### Terminal (brief summary, 15-20 lines)\n\n```\n**Review Complete** — uncommitted\n**Verdict**: PASS WITH CONCERNS\n**Findings**: 4 issues (confidence >= 80) | 1 consensus\n\n**Critical Issues** (1):\n- [P1] [Confidence: 92] Missing null check on user input — src/api/users.ts:42\n\n**Top Recommendations**:\n1. Add input validation before database query\n2. Add test for empty user ID case\n\n**Detailed Report**: /tmp/gemini-review/20260214-153022/summary.md\n```\n\n### Detailed reports (written to /tmp/)\n\n```\n/tmp/gemini-review/{timestamp}/\n├── summary.md        # All findings with evidence\n├── security.md       # Security reviewer output\n├── correctness.md    # Correctness reviewer output\n└── quality.md        # Quality reviewer output\n```\n\n## Architecture\n\n```\n/review command (TOML)\n    │\n    ├── Injects diff via !{git diff HEAD}\n    │\n    └── Gemini main agent (guided by GEMINI.md)\n         │\n         ├── Delegates to security-reviewer (sub-agent)\n         ├── Delegates to correctness-reviewer (sub-agent)\n         ├── Delegates to quality-reviewer (sub-agent)\n         │\n         ├── Synthesizes: de-duplicate → consensus bonus → threshold → sort\n         ├── Writes detailed reports to /tmp/gemini-review/\n         └── Returns brief terminal summary\n```\n",
  "bytes": 4513,
  "sha": "b9dbcc672fcfb1dc836fa436c57caf1fccc3605d3104a747a15f1ed8f8b294a0",
  "repo_slug": "shravansunder/gemini-review-agents",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_shravansunder_gemini_review_agents_8bc909b8/readme"
}