release
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
gh(GitHub CLI),git, andpnpmto interact with the repository and CI/CD workflows. These commands are integral to the release process and target the official repository of the vendor. - [INDIRECT_PROMPT_INJECTION]: The skill ingests data from GitHub pull requests to identify potential regressions or fixes that should be included in a release. This data is untrusted as it can be submitted by external contributors.
- Ingestion points: The
gh pr listcommands inSKILL.mdretrieve titles and bodies of open pull requests. - Boundary markers: The skill does not provide specific instructions or delimiters to the agent to distinguish between the fetched PR data and the skill's own instructions.
- Capability inventory: The agent has the capability to trigger production releases (
gh workflow run), push git tags (git push), and execute build scripts (pnpm release:cut). - Sanitization: While the scripts use
jqto extract metadata, the text content interpreted by the AI is not sanitized to prevent embedded instructions from influencing the agent's release decisions.
Audit Metadata