project-guide-validation
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute several shell commands for testing and environment setup.\n
- Evidence: Use of
node --test src/authoring/guides-integrity.test.tsfor automated validation.\n - Evidence: Use of
pnpm buildandpnpm baselinestatusfor build and status checks.\n - Evidence: Use of
git show <base_sha>:<path>to inspect files during PR reviews.\n - Evidence: Instructions to start a local server using
python3 -m http.server 8080ornpx http-servervia arun_commandtool.\n - Evidence: Use of custom project CLI tools like
gd devandgd baselinestatus.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external, potentially untrusted content during the validation workflow.\n - Ingestion points: Reads and analyzes
guide.md,demo.html, andexpectations.mdfiles (SKILL.md), as well as git diffs and PR comments.\n - Boundary markers: The skill does not explicitly define clear delimiters or "ignore instructions" wrappers for the content being validated.\n
- Capability inventory: The agent has access to shell command execution (
run_command), file system reading, and browser automation via DevTools MCP (click,type_text, etc.).\n - Sanitization: No explicit sanitization or filtering of the input files is described before they are processed by the agent or loaded into the browser.\n
- Risk: Malicious instructions embedded in the documentation or demos could attempt to manipulate the agent's validation report or exploit the browser automation tools.
Audit Metadata