agentic-workflows
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches workflow prompts and configuration files from the official github/gh-aw repository using curl. These downloads are used as a fallback for the GitHub API and target a well-known, trusted organization repository.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes data from Markdown files in the local repository and the guidance repository to drive agent behavior.
- Ingestion points: Reads project-specific files such as .github/aw/instructions.md and various workflow-specific prompts (e.g., create-agentic-workflow.md).
- Boundary markers: The instructions do not define explicit boundary markers or delimiters for the loaded content, instead directing the agent to "follow it directly."
- Capability inventory: The agent maintains standard capabilities to perform the routed tasks, including file system access and network operations via curl.
- Sanitization: No explicit sanitization or schema validation of the loaded Markdown content is performed before the instructions are followed.
Audit Metadata