write-mcp-test
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill requires the installation of
@gleanwork/mcp-server-tester, which is a package provided by the skill author ('gleanwork'). It also utilizes standard testing and validation libraries including@playwright/testandzod. - [INDIRECT_PROMPT_INJECTION]: The skill defines a testing framework where untrusted data (tool outputs from an MCP server) is ingested and evaluated by an LLM.
- Ingestion points: Data enters the agent's context through the
mcp.callToolfunction, which is then passed to matchers liketoContainToolText,toMatchToolPattern, and notablytoPassToolJudge. - Boundary markers: None are present in the provided templates to isolate the tool output from the evaluation instructions.
- Capability inventory: The generated tests are executed via
npx playwright test, which involves shell command execution to run the test suite. - Sanitization: The skill provides instructions for sanitizing snapshots (e.g., removing UUIDs or timestamps) but does not include sanitization for tool outputs before they are processed by the LLM judge matcher (
toPassToolJudge).
Audit Metadata