Threat Model (summary)

• Trust boundaries: GitHub Issue authors (untrusted), CI runners (trusted), GitHub API (trusted with scoped tokens).
• Data exposure: do not leak secrets in comments; redact tokens.
• Attack vectors: large issue bodies (limit 1 MB), malicious YAML (use yaml.safe_load), rate limiting / abuse (backoff).
• Mitigations: minimal token scopes, safe YAML loading, idempotent comments with hidden markers, artifact-only run history.