fix-failing-pipelines
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub Actions build logs to determine failure causes and generate fix instructions for another tool.
- Ingestion points: Workflow logs are retrieved via
gh run view --log-failedin Step 3. - Boundary markers: The instructions do not define delimiters or warnings to ignore instructions that might be embedded within the log content.
- Capability inventory: The skill uses the
ghCLI to interact with repositories and invokes the/babysitter:yoloskill for automated code changes and commits. - Sanitization: There is no evidence of filtering, escaping, or sanitizing the log data before it is interpolated into the prompt for the next stage.
- [COMMAND_EXECUTION]: The skill uses the
ghCLI to list workflow runs, view failure details, and check the status of new runs. While these are standard development operations, they provide the necessary surface for interacting with external data sources that could influence agent behavior.
Audit Metadata