The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external YAML-based definitions for tests, suites, and hooks, which represents an ingestion surface for potential instructions designed to override agent behavior.
Ingestion points: Test and suite definitions are read from the local workspace files by the MCP tools and CLI fallbacks as described in SKILL.md.
Boundary markers: The instructions in SKILL.md require the agent to 'inspect the requested scope' and 'obtain explicit confirmation' from the user before running tests or performing file mutations.
Capability inventory: The skill workflow in SKILL.md includes capabilities to modify files in the workspace and execute shell commands through npx and the agent-qa CLI.
Sanitization: All definitions are required to undergo validation using the agent_qa_validate_test tool or the agent-qa ids validate command prior to being saved or enqueued for execution.
[COMMAND_EXECUTION]: The skill uses npx --yes agent-qa as a fallback mechanism for validating definitions and generating identifiers, which involves fetching and executing code from the npm registry.

agent-qa-authoring