relay-plan
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from GitHub issues, local task files, and user descriptions to generate planning artifacts. This creates a surface where malicious instructions embedded in the source data could influence the agent's logic during the planning phase or the content of the generated dispatch prompt for downstream agents.
- Ingestion points: External content is ingested via 'gh issue view' and local file reads from 'backlog/tasks/' and '~/.relay/requests/'.
- Boundary markers: While the skill refers to a prompt template, it does not explicitly implement delimiters or safety instructions within the provided files to isolate untrusted external text.
- Capability inventory: The skill can execute local environment-probing scripts and file-persistence scripts, and it generates the primary input prompt for the subsequent 'relay-dispatch' agent.
- Sanitization: No sanitization or filtering of the external task descriptions was observed in the provided logic.
Audit Metadata