The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes Node.js helper scripts (playwright-lifecycle.mjs and server-probe.mjs) which execute shell commands provided via arguments (such as --setup, --server, and --run). These commands are executed using node's child_process.spawn with a shell (bash -lc), providing a high level of control over the environment.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it is designed to ingest and process untrusted data from multiple sources. 1. Ingestion points: External data enters the agent's context through files in the .tasks/ directory and AGENTS.md. 2. Boundary markers: The workflow uses Markdown tables and dedicated artifact files to structure data, and includes explicit instructions to treat input files as read-only. 3. Capability inventory: The agent has capabilities for file system operations (rm, mkdir, cp), file inspection (sed, rg, ls), and shell command execution via the provided lifecycle scripts. 4. Sanitization: The skill emphasizes manual review and artifact-driven gates but does not specify automated sanitization or escaping of the ingested data before it is referenced in commands.\n- [SAFE]: The skill includes several security-conscious features, such as a blocklist in the lifecycle helper script to prevent the modification of critical database environment variables and a prohibition against using the production database for testing.

task-workflow