qc-optimizer
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes project source code and can execute shell commands, creating a potential vector for indirect prompt injection.
- Ingestion points: The skill uses codex_read_file and codebase_search tools to read the entire project repository (SKILL.md, agents/openai.yaml).
- Boundary markers: The agent's system prompt does not include instructions to differentiate between user instructions and instructions embedded in the data being analyzed (agents/openai.yaml).
- Capability inventory: The agent has access to shell tools including bash, npm, and cargo, which can be leveraged if an injection is successful (agents/openai.yaml, SKILL.md).
- Sanitization: No filtering or sanitization of input read from files was identified.
Audit Metadata