feature-delivery
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill instructs the agent to execute external scripts (config.mjs, relay.mjs) from other skill directories (delegate-setup, *-delegate) using the node command. This involves runtime execution of code that is not contained within this skill. However, these appear to be coordinated components within the agent ecosystem.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local repository to generate execution plans and task briefs.
- Ingestion points: Reads AGENTS.md, repository status, diffs, architecture documents, and source code in SKILL.md (Section 1).
- Boundary markers: Uses a structured XML format for task briefs as defined in references/task-brief-template.md.
- Capability inventory: Authorized to execute shell commands via node and read/write repository files.
- Sanitization: No specific sanitization of the content read from repository files is documented before it is processed.
Audit Metadata