shepherd
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes 'gh' CLI commands to retrieve pull request data and post automated replies. These operations are restricted to the GitHub platform and are essential to the skill's triage and review workflow.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of pull request threads and bot findings.
- Ingestion points: Untrusted content is ingested from GitHub pull request comments via the 'gh' CLI as described in the 'Wait for the round' section.
- Boundary markers: The instructions mandate that the agent must always read the local specification file ('docs/specs/*.md') and explicitly avoid using the pull request's summary as a source of truth.
- Capability inventory: The skill can post replies to GitHub threads and trigger external build processes through the '/build' and '/crucible' commands.
- Sanitization: The skill requires operator 'action' escalation before replying to colleagues and 'decision' escalation for any findings that suggest changes to the specification.
Audit Metadata