project-memory
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for a 'Pinned repository ledger' (
references/repo-ledger.md) where agents are instructed to read, parse, and act upon status updates or commands posted as comments in a GitHub issue. This creates a channel where external content can influence agent behavior. - Ingestion points: The agent is directed to ingest instructions from repository files (
AGENTS.md,CLAUDE.md) and a designated pinned GitHub issue ledger. - Boundary markers: The ledger mechanism uses HTML comment delimiters (
<!-- ledger-status:start -->) and specific command prefixes (To <agent>:) to identify relevant content. - Capability inventory: The skill involves file writing (creating/updating instruction files) and interacting with the GitHub API (reading issues and checking collaborator permissions).
- Sanitization: The skill provides critical safety guidance by requiring the agent to verify that the posting account has appropriate repository permissions ('admin', 'maintain', or 'write') via the GitHub API before treating any comment as a valid instruction.
Audit Metadata