stitch-loop
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The autonomous 'baton' loop relies on instructions passed via the .stitch/next-prompt.md file, where each iteration prepares the task for the next. This creates a functional surface where input from one step influences subsequent actions.
- Ingestion points: The agent reads task details from .stitch/next-prompt.md and project context from .stitch/SITE.md.
- Boundary markers: There are no explicit markers or safety instructions directing the agent to validate or delimit the contents of the baton file.
- Capability inventory: The agent uses tools for code generation, filesystem writes, and shell command execution.
- Sanitization: The skill does not perform sanitization on the file paths or the instruction text before execution.
- [COMMAND_EXECUTION]: The optional visual verification step uses the Bash tool to execute 'npx serve' for hosting the generated site locally.
- [EXTERNAL_DOWNLOADS]: The skill fetches generated code and images from Stitch service URLs and downloads the 'serve' package from the npm registry via npx.
Audit Metadata