Security review: API keys, token handling, permissions model

[nev-offload]

Problem

Before public release, the action needs a security review of how it handles sensitive data and permissions.

Areas to review

1. API key handling

   • Keys are passed as GitHub Secrets (good) but need to verify they're never logged
   • Mask all sensitive values in action output (core.setSecret())

2. GitHub token permissions

   • Document minimum required permissions
   • Use least-privilege: only request issues: write, pull-requests: write, contents: read
   • Support fine-grained PATs

3. Agent sandboxing

   • What tools does the agent have access to?
   • Can it execute arbitrary code? If so, in what context?
   • Should there be an allowlist of permitted operations?

4. Input validation

   • Sanitize issue/PR content before passing to the agent (prompt injection risk)
   • Don't let issue authors trick the agent into revealing secrets or performing unintended actions

5. Rate limiting

   • Prevent abuse (e.g., someone opening 100 issues to burn API credits)
   • Consider per-repo or per-user rate limits

Priority

P1 — Must be addressed before any public/marketplace release.