The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a Python script (validate_schema.py) to validate stage outputs and schema consistency. It also orchestrates the execution of automated test scripts (Playwright or API tests) residing in the project's repository during the execution phase.
[DYNAMIC_EXECUTION]: In Stage 5 (Execution), the skill triggers the running of external scripts or code found within the user-provided code repository. While this is the primary purpose of the skill, it represents a vector for executing code provided in the input context.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data including requirement materials and code repositories as defined in Workflow Step 1. This data is processed through multiple reasoning stages (Requirement Analysis, Test Strategy, Case Writing), creating a surface where hidden instructions in the documents could influence the agent's orchestration logic.
Ingestion points: Requirements documents, code repositories, and user clarification records provided at the start of the workflow.
Boundary markers: The skill defines stage-level context isolation and requires human confirmation at multiple checkpoints (⏸) to review proposals and clarify ambiguities.
Capability inventory: Execution of validation scripts (validate_schema.py), execution of automated test suites (Stage 5), and file writing across the project directory.
Sanitization: No explicit sanitization of document content is described, relying instead on the structural isolation of sub-sessions and human review.
[PERSISTENCE]: The skill implements a 'cross-session knowledge' mechanism that writes findings such as environment quirks or flaky test patterns into a local .qa/ directory via the qa-memory function. This allows information to persist and influence the behavior of the agent in subsequent sessions.

qa