pdd-check
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes the
pddcommand-line tool via instructions likeCI=1 pdd test <story> --cache-clear --jsonandpdd workflow status N. These commands are used to perform testing and update status on the platform. - [DYNAMIC_EXECUTION]: The instructions direct the agent to "Build the head" and "run the changed path with the repo recipe." This involves executing the build system (e.g., npm, make, pip) and running the application code directly from the pull request being analyzed. This is a high-capability action that relies on the integrity of the repository content.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted data from pull requests (issues, stories, and code diffs).
- Ingestion points: The agent reads issue descriptions, PR stories, and code diffs from the head of a branch.
- Boundary markers: The skill attempts to implement a trust boundary with the instruction "Read only trusted people's text (pdd-autopilot)," though this relies on the agent's ability to distinguish authors within the provided context.
- Capability inventory: The skill has the capability to execute shell commands (
pdd), initiate build processes, and run arbitrary code paths from the repository. - Sanitization: There is no mention of sanitization or isolation (such as containers) for the execution of the repo's build recipe or changed paths.
Audit Metadata