The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines standardized procedures for interacting with the host environment to manage project lifecycles. This includes discovering build and test commands from project manifests, starting development servers in the background using tool-specific mechanisms or shell commands (like nohup), and performing network health checks via curl on localhost. It also includes comprehensive cleanup protocols that utilize process management tools (pgrep, kill, pkill) to terminate only those process trees initiated by the agent.
[INDIRECT_PROMPT_INJECTION]: The skill manages the ingestion of untrusted project data, such as configuration files and task definitions. It incorporates advanced security 'guards' into its procedures, specifically instructing the agent to never read full configuration files that might contain secrets, to filter sensitive keywords (e.g., 'token', 'secret', 'password') before processing content, and to suppress detailed error messages that might inadvertently leak file snippets to the model's context.
[SAFE]: The skill's primary focus is establishing safe operational boundaries. It emphasizes the principle of least privilege by forbidding global tool installations, discouraging the use of sudo, and ensuring that background services are only started when necessary and stopped immediately after use.

do-shared