ce-work
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes verification commands defined within the input plan documents (specifically the 'Verification Contract'). These commands are used to validate code changes before they are committed to the repository.
- [EXTERNAL_DOWNLOADS]: The skill downloads 'Compound Packs' (rule sets) from external Git repositories specified in the project configuration. It also uses 'npx' to fetch the 'acpx' agent transport package from the NPM registry.
- [DATA_EXFILTRATION]: Repository content is sent to external AI model providers (such as Anthropic, OpenAI, and Google) to perform implementation tasks. The skill implements a redaction mechanism to prevent the accidental transmission of secrets during these sessions.
- [INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data from plans and prompts. It includes explicit instructions to treat external rule text as evidence rather than direct instructions, which helps mitigate potential injection attacks.
- [DYNAMIC_EXECUTION]: The skill dynamically assembles and executes implementation and verification logic based on the contents of the plan provided at runtime.
Audit Metadata