codex-build-loop
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill reads task definitions and roadmaps from files within the repository (e.g.,
docs/ROADMAP.md,docs/REFACTOR.md, anddocs/SECURITY-AUDIT.md). If these files are influenced by an untrusted source, such as a malicious pull request, they could contain instructions that redirect the agent's behavior. - Ingestion points: Instructions are retrieved from
docs/ROADMAP.md,docs/REFACTOR.md, anddocs/SECURITY-AUDIT.mdto determine the tasks the agent will perform. - Boundary markers: The instructions do not define delimiters or provide specific prompts to ensure the agent treats the file content as data rather than instructions.
- Capability inventory: The agent is empowered to modify the codebase, run test suites (executing code), and invoke the
/reviewtool based on the ingested content. - Sanitization: There is no specified verification or sanitization process for the logic described in the task files before the agent implements the changes.
Audit Metadata