lets-go
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to follow instructions provided in external handoff documents which could be manipulated.
- Ingestion points: The agent reads 'Next steps' and 'Machine-local knowledge' from handoff files (e.g.,
docs/agents/handoffs.mdand related documents in a Handoffs folder). - Boundary markers: No specific delimiters or warnings to disregard embedded instructions are present in the skill body.
- Capability inventory: The skill performs shell operations including git commands, node script execution, and dependency installation (
npm ci,pnpm install). - Sanitization: While the skill notes that actions gated by
AGENTS.mdrequire user confirmation, it does not explicitly sanitize the text of the 'Next steps' before processing them. - [COMMAND_EXECUTION]: The skill executes several shell commands to manage the repository state.
- Evidence: Execution of
node .agents/skills/handoff/handoff-map.mjs,git fetch origin --prune, and automated dependency installation from lockfiles. - [DYNAMIC_EXECUTION]: The skill retrieves and executes shell commands defined in repository-local markdown files.
- Evidence: Settings like 'Workspace status command', 'Sync with main', and 'Companion restore command' are read from
docs/agents/workspace.mdand executed to bring the environment to the handoff state.
Audit Metadata