handoff-note
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFE
Full Analysis
- [DATA_EXPOSURE]: The skill includes rigorous instructions for the redaction of secrets, credentials, tokens, and private identifiers (PII) before generating any output. It specifically requires redacting local usernames, home directories, and private machine identifiers unless they are strictly necessary for the work to continue.
- [PROMPT_INJECTION]: The skill implements a defensive strategy against prompt injection by instructing the agent to treat source material (such as logs, issue text, and PR descriptions) as 'evidence' rather than 'authority.' It warns the agent not to follow instructions embedded within this data unless they are explicitly endorsed by the user.
- [INDIRECT_PROMPT_INJECTION]: The skill identifies and mitigates risks associated with processing untrusted workspace data.
- Ingestion points: Workspace files, git diffs, command logs, and issue text (specified in SKILL.md).
- Boundary markers: Present; the instructions require the agent to quote untrusted content or place it under a specific 'Open Questions' section to prevent it from being interpreted as a system requirement.
- Capability inventory: Reading repository state, file contents, and generating markdown documents. No high-privilege operations or network exfiltration patterns are present.
- Sanitization: Explicit redaction of sensitive data and a requirement to verify all assertions against workspace evidence.
Audit Metadata