The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because the agent is directed to ingest and act upon test execution results, including failure diffs, stack traces, and console logs.
Ingestion points: The agent reads failure details and console output to diagnose test regressions (found in 'references/running-as-an-agent.md').
Boundary markers: The skill recommends the 'minimal' reporter to reduce the volume of passing-test output, though it does not provide explicit prompt delimiters for the ingested log content.
Capability inventory: The agent is authorized to execute shell commands ('vitest run') and modify project files (configurations, source code, and snapshots).
Sanitization: There are no instructions for the agent to sanitize or escape the test output before it is processed as part of the conversation context.
[EXTERNAL_DOWNLOADS]: The skill mentions third-party extensions and provider packages required for specific Vitest features.
References third-party MCP servers such as '@madrus/vitest-mcp-server' and 'djankies/vitest-mcp' in 'references/running-as-an-agent.md'.
Includes a mandatory security instruction that human approval MUST be obtained before adding these third-party tools to a toolchain.
[COMMAND_EXECUTION]: The skill guides the agent in the operational use of the Vitest command-line interface.
Provides specific instructions for running 'vitest run' with various flags for filtering by path, name, or line number to optimize execution and output ('references/running-as-an-agent.md').
Includes diagnostic commands such as 'vitest list' and debugging flags like '--inspect-brk' ('references/debugging-and-failure-modes.md').

vitest-testing