The Agent Skills Directory

[SAFE]: The skill implements proactive security measures, requiring that all user-provided configuration be treated as untrusted data and forbidding the agent from performing automated installation or download tasks. It provides clear warnings regarding the use of 'replace' mode on shared clusters to prevent unintended service disruption.\n- [COMMAND_EXECUTION]: The skill assists in generating and running local development commands (e.g., mirrord up, node). It enforces human review for all commands and specifically mandates regex validation for Kubernetes names to prevent shell injection.\n- [PROMPT_INJECTION]: The skill addresses potential indirect injection risks by limiting the scope of Tera templating to only key and get_env functions, and by establishing clear boundaries for processing external mirrord-up.yaml files. Mandatory evidence chain: 1. Ingestion points: mirrord-up.yaml configuration provided by the user. 2. Boundary markers: Explicit instructions to treat user data as untrusted and follow defined security rules. 3. Capability inventory: Local command execution via mirrord and specified runtimes (e.g., node, python). 4. Sanitization: Regex-based validation of workload names and namespaces before interpolation into shell commands.

mirrord-up