The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute local shell commands to run project tests (e.g., npm, pytest). It contains extensive instructions in references/execution.md to verify commands, use isolated test data, and avoid expanding access.
[DYNAMIC_EXECUTION]: The skill facilitates the generation and execution of test scripts at runtime, particularly for browser automation.
Evidence: references/playwright-e2e.md describes coordinating "AI-generated Playwright journeys" and executing "native test commands" based on these generated scripts.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection because it processes untrusted external data and has the capability to execute code.
Ingestion points: Reads local source code (references/workflow.md), API responses (references/api-and-logic.md), and DOM/browser content (references/page-walkthrough.md).
Boundary markers: Implements structured result contracts (references/result-contract.md) and mandates explicit labeling of results (e.g., "Mock", "Real", "Inferred") to maintain data isolation.
Capability inventory: Full capability to execute shell commands for testing purposes and browser automation tools like Playwright as detailed in references/execution.md and references/playwright-e2e.md.
Sanitization: Instructions explicitly mandate redacting tokens, cookies, and personal data from logs, and scrubbing credentials from durable evidence (references/api-and-logic.md and references/execution.md).

dev-tests