The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Accesses sensitive files to retrieve application secrets and authentication data.
Evidence: Instructs the agent to read .env.local, .env.test, and README files to identify credentials for the application under test.
[COMMAND_EXECUTION]: Performs arbitrary command and code execution through shell and MCP tools.
Evidence: Executes shell commands including git, grep, and npm to analyze the repository and manage development servers.
Evidence: Utilizes browser_run_code_unsafe and browser_evaluate via the Playwright MCP to interact with the browser context.
Evidence: Leverages the Supabase MCP to execute raw SQL queries (execute_sql) and apply database migrations (apply_migration).
[PROMPT_INJECTION]: Susceptible to indirect prompt injection through untrusted data ingestion.
Ingestion points: browser_snapshot (SKILL.md) and firecrawl_search (SKILL.md) provide external content that the agent must process.
Boundary markers: Absent in the workflow instructions for delimiting tool outputs.
Capability inventory: Shell execution, execute_sql, browser_run_code_unsafe, and the ability to perform full-stack file modifications.
Sanitization: No sanitization, validation, or escaping steps are defined for data retrieved from browser sessions or web search results.

test-playwright