github-actions-integration
Pass
Audited by Gen Agent Trust Hub on Jun 22, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: Potential indirect prompt injection surface in the workflow templates provided for issue resolution.
- Ingestion points: The
.github/workflows/claude-fix.ymlfile ingests untrusted data from${{ github.event.issue.title }}and${{ github.event.issue.body }}directly into the agent's instructions. - Boundary markers: The prompt template lacks delimiters or explicit instructions to distinguish between agent commands and external data content.
- Capability inventory: The workflows request
contents: writeandpull-requests: writepermissions, allowing the agent to modify code and create PRs based on instructions derived from potentially malicious issues. - Sanitization: No sanitization, escaping, or validation of the issue title or body is performed before interpolation into the prompt.
- [EXTERNAL_DOWNLOADS]: The skill references and utilizes the
anthropics/claude-code-action@v1GitHub Action from a well-known source to facilitate its automated tasks.
Audit Metadata