manki-review/Manki Review
Multi-agent AI code review for GitHub. Planner picks the team, agents review in parallel, judge filters the noise.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Stale
- License
- None
Inputs
| name | description | required | default |
|---|---|---|---|
| claude_code_oauth_token | Claude Code OAuth token for Max subscription authentication | no | — |
| anthropic_api_key | Anthropic API key (alternative to OAuth token) | no | — |
| openai_oauth_token | Base64-encoded contents of `~/.codex/auth.json` (Codex CLI OAuth credentials, including `tokens.access_token` and `tokens.refresh_token`). Bootstrap with `cat ~/.codex/auth.json | base64 | gh secret set OPENAI_OAUTH_TOKEN`. Re-run when the refresh_token expires. | no | — |
| openai_api_key | OpenAI API key (alternative to OAuth token) | no | — |
| gemini_oauth_token | Base64-encoded contents of `~/.gemini/oauth_creds.json` (Gemini CLI OAuth credentials, including `access_token` and `refresh_token`). Bootstrap with `cat ~/.gemini/oauth_creds.json | base64 | gh secret set GEMINI_OAUTH_TOKEN`. Re-run when the refresh_token expires. | no | — |
| gemini_api_key | Google Generative AI API key (alternative to Gemini OAuth token) | no | — |
| github_token | GitHub token for posting reviews and comments (not required when using GitHub App auth) | no | — |
| config_path | Path to .manki.yml config file | no | .manki.yml |
| memory_repo_token | GitHub token with write access to the review memory repository (patterns, learnings, suppressions) | no | — |
| github_app_id | GitHub App ID for custom bot identity (reviews show as the app name) | no | — |
| github_app_private_key | GitHub App private key (PEM format) for generating installation tokens | no | — |
| manki_token_url | Token service URL for GitHub App identity (default: https://manki-api.dustinface.me/token) | no | https://manki-api.dustinface.me/token |
| concurrency_lock_ttl_seconds | TTL (in seconds) for the in-app concurrent-submission lock. If another `manki-review[bot]` run has posted an "in progress" marker comment within this window, the current run bails before any LLM call. Acts as defense in depth on top of the workflow-level concurrency group. Default: 600 (10 minutes). | no | 600 |
Outputs
| name | description |
|---|---|
| review_id | The GitHub review ID that was posted |
| verdict | Review verdict: APPROVE, COMMENT, or REQUEST_CHANGES |
| findings_count | Total number of findings |
| findings_json | JSON array of all findings (for downstream processing) |
| severity_counts | JSON object with severity counts: {blocker, warning, suggestion, nitpick} |
| judge_model | Model used for the judge agent |