pr-design-doc
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes standard command-line tools including
git,gh(GitHub CLI),mkdir, andtestto perform repository operations such as checking out PR branches, comparing code versions, and creating a temporary directory for artifacts. These operations are integral to the skill's purpose and include validation steps like SHA verification and symlink detection. - [INDIRECT_PROMPT_INJECTION]: The skill ingests potentially untrusted data from repository pull requests (code, titles, and descriptions). It mitigates the risk of this content influencing the agent or compromising the reviewer's browser by enforcing strict HTML escaping and sanitization of all source-derived values before they are inserted into the generated design document, as outlined in the
references/html-craft.mdfile. - [DATA_EXPOSURE]: The skill manages visibility risks by providing specific workflows for public and private repositories. It advises using the
htmlpreview.github.ioservice for public repos while explicitly forbidding its use for private or internal repositories to prevent data leakage, recommending local previews or access-controlled services instead.
Audit Metadata