cicd-integration
Pass
Audited by Gen Agent Trust Hub on Apr 12, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill documents the use of official GitHub Actions from the 'anthropics' organization and the '@anthropic-ai/claude-code' npm package, which are trusted resources.- [EXTERNAL_DOWNLOADS]: References standard and well-known GitHub Actions including 'actions/checkout' and 'actions/github-script' for workflow automation.- [DATA_EXFILTRATION]: Contains patterns for communicating with the GitHub API to automate pull request comments, which is an expected behavior for CI/CD tooling.- [COMMAND_EXECUTION]: Provides examples of shell scripts and git hooks using the 'Bash' tool to perform repository operations, file checks, and build validations.- [PROMPT_INJECTION]: The skill describes workflows that ingest untrusted data from pull request titles and file contents, creating an attack surface for indirect prompt injection.
- Ingestion points: PR titles, commit messages, and repository source code (documented in SKILL.md).
- Boundary markers: None identified in the provided example prompt templates.
- Capability inventory: Access to 'Read', 'Grep', 'Glob', 'Bash', and 'Write' tools (documented in SKILL.md).
- Sanitization: No explicit sanitization or filtering of external inputs is shown in the integration examples.
Audit Metadata