The Agent Skills Directory

[COMMAND_EXECUTION]: The skill discovers and executes arbitrary shell commands found within a repository's documentation, CI configurations, or build manifests (e.g., npm test, pytest, make test).
Evidence: SKILL.md describes identifying and running 'primary test entry points' and 'CI workflow steps' parsed from repository files.
[EXTERNAL_DOWNLOADS]: The skill performs dependency installations and network operations if the repository's test suite requires them.
Evidence: SKILL.md includes logic for npm ci, docker compose, and environment setup, though it mandates user confirmation before proceeding.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it parses untrusted repository content to determine its next actions.
Ingestion points: Reads README.md, package.json, .github/workflows/*.yml, and other manifest files during the 'Discover test plan' phase.
Boundary markers: The skill relies on manual user confirmation as a safety boundary before executing discovered commands, but lacks automated technical delimiters for the parsed data.
Capability inventory: The skill possesses shell execution capabilities (subprocess calls) across all scripts to run tests and install dependencies.
Sanitization: No explicit sanitization or structural validation is performed on the commands extracted from documentation or manifests before they are presented to the user for execution.

automate-tests