create-agent-harness
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill performs an evidence-based discovery process by analyzing the contents of a target repository (directory structure, tech stack, conventions). This ingestion of potentially attacker-controlled data creates a surface where malicious files in the target repo could attempt to influence the harness generation or the agent's behavior.
- Ingestion points: Target repository files, directory structure, and existing configuration files discovered during the discovery phase.
- Boundary markers: The instruction 'Forbidden: invent context' acts as a logical constraint, but the skill lacks technical delimiters to isolate the discovered repository data from the agent's core instructions.
- Capability inventory: The skill directs the agent to create numerous configuration files and directory structures at the repository root, as well as define sub-agents with specific tools (Read, Grep, Glob).
- Sanitization: There is no evidence of sanitization or filtering of the information extracted from the target repository before it is used to populate harness templates.
Audit Metadata