The Agent Skills Directory

[SAFE]: The skill provides legitimate developer workflows for agent engineering and follows security best practices for prompt engineering and environment management.
[COMMAND_EXECUTION]: The skill scaffolds utility bash scripts for tasks like link checking and log processing. It also provides command templates for running agents in isolated sandboxes using standard claude CLI flags like --safe-mode and --allowedTools.
[EXTERNAL_DOWNLOADS]: Mentions well-known and trusted external tools such as Vercel's agent-browser and Playwright for UI testing. These references are documented neutrally as recommendations for end-to-end testing and do not involve unverified third-party sources.
[INDIRECT_PROMPT_INJECTION]: While the skill processes user-provided descriptions of agent failures (untrusted input), it mitigates risks by recommending containment techniques, explicit schema validation, and dry-running generated artifacts in restricted environments to verify their behavior.

pick-harness