nfs-247/AI Peer Review Dispatcher
Runs the AI peer review dispatcher against the calling repository's pull request. Claude/GPT/Gemini review adversarially and post HMAC-signed verdicts; the dispatcher converges or escalates to the human operator. It never merges (no contents: write in the caller).
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Stale
- License
- None
Inputs
| name | description | required | default |
|---|---|---|---|
| github_token | Token for GitHub API calls. Pass secrets.GITHUB_TOKEN. | yes | — |
| anthropic_api_key | Anthropic API key (Claude reviewer). | no | "" |
| openai_api_key | OpenAI API key (GPT reviewer). | no | "" |
| gemini_api_key | Gemini API key (Gemini reviewer, high-stakes). | no | "" |
| xai_api_key | xAI API key (optional Grok reviewer). Only called when 'grok' is added to a tier roster; an unset key is harmless. | no | "" |
| resend_api_key | Resend API key for escalation email (optional; falls back to a PR comment). | no | "" |
| operator_email | Escalation email recipient (optional). | no | "" |
| operator_github_login | GitHub login allowed to issue OPERATOR commands. | no | NERT24 |
| dispatcher_verdict_secret | HMAC key signing dispatcher verdicts. Without it, no verdict counts (fail-safe). | no | "" |
| google_chat_webhook_url | Google Chat incoming-webhook URL for mobile escalation pings (optional). | no | "" |
| approve_webapp_url | Apps Script /exec URL for one-tap operator approval from the Chat card (optional). | no | "" |
| approve_signing_secret | Shared secret to HMAC-sign approve/merge links so they can't be edited to target other PRs (optional). | no | "" |
| repo_config_path | Path (relative to the caller checkout) to the per-repo config. Defaults to .peer-review.json at the repo root; the legacy .github/ai-peer-review.json is still loaded automatically if the primary file is absent. | no | .peer-review.json |
| escalation_cooldown_minutes | Minutes the dev agent must be quiet (no new commit) before a review-outcome escalation pings the operator. Empty uses the repo config value (default 10). Set 0 to ping immediately. | no | "" |
| billing_mode | How this tenant is billed for AI usage: 'byok' (tenant's own keys, platform charges nothing for usage) or 'platform' (platform's keys, billed with a markup). Empty uses the repo config value (default byok). | no | "" |
| usage_markup_multiplier | Multiplier applied to provider cost when billing_mode=platform (e.g. 1.3 = 30% margin). Empty uses the repo config value (default 1.0). | no | "" |
| dev_fee_usd | Flat per-period dev/orchestration fee recorded for invoicing. Empty uses the repo config value (default 0). | no | "" |
| app_id | GitHub App id. When set with app_private_key, the dispatcher mints a per-installation token (its own 5–15k req/hr quota) instead of the per-repo GITHUB_TOKEN (1k/hr). Optional; see SCALING.md. | no | "" |
| app_private_key | PEM private key for the GitHub App in app_id. Pass secrets.APP_PRIVATE_KEY. Required only if app_id is set. | no | "" |
Outputs
no outputs