development-harness
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from feature descriptions and external backlog providers, creating a potential surface for indirect prompt injection to influence downstream automated workflows.\n
- Ingestion points: User-provided strings for
{feature description}and external task data retrieved from the backlog backend (GitHub Issues) are ingested into the routing process as defined inSKILL.md.\n - Boundary markers: No specific use of delimiters or 'ignore' instructions are provided in the instruction set to isolate untrusted content from the agent's logic.\n
- Capability inventory: The skill coordinates other tools that perform code implementation, agent dispatching, and repository management (e.g., committing and pushing code).\n
- Sanitization: The provided files do not describe any sanitization, validation, or filtering of the ingested external data.
Audit Metadata