The Agent Skills Directory

[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by defining workflows that ingest untrusted fixture data from the filesystem and pass it to automated tools such as OCR and security scanners.
Ingestion points: Files in the tests/fixtures/ directory including images, PDFs, and text files.
Boundary markers: None implemented in the provided test templates to isolate untrusted content from instructions.
Capability inventory: The skill manages MCP server lifecycles as child processes and provides a callTool interface to interact with filesystem and data tools.
Sanitization: No evidence of sanitization or validation of the input fixture data prior to tool processing.
[COMMAND_EXECUTION]: The TestHarness utility is described as starting and stopping MCP servers as child processes and executing the test runner via npx vitest.
[DATA_EXFILTRATION]: The skill includes instructions for testing security tools by scanning sensitive local files, such as .env files and documents containing simulated PII, for reconciliation and auditing purposes.
[EXTERNAL_DOWNLOADS]: The documentation suggests using vitest, a well-known technology tool, which may be fetched and executed through the npx package runner.

tool-chain-test