The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages the project development lifecycle by executing numerous git commands, worktree operations, and custom CLI tools. It includes a synchronization script (sync_configuration.py) that performs extensive file system operations, including reading local project workflows and rules, and removing legacy command directories.
[EXTERNAL_DOWNLOADS]: The skill installs third-party Python packages spec-kitty-cli and gemini-agent via pip to provide core functionality. These packages are related to the skill's stated purpose and well-known services.
[PROMPT_INJECTION]: The skill utilizes strict 'Anti-Simulation Rules' to override default agent behavior and enforce the execution of commands. It also presents an indirect prompt injection surface by transforming natural language requirements into implementation tasks.
Ingestion points: Processes untrusted content from spec.md, plan.md, and user-provided $ARGUMENTS in various templates.
Boundary markers: Relies on standard markdown headers but lacks explicit 'ignore instructions' delimiters for interpolated user requirements.
Capability inventory: Full access to git repository management, file system modification via shutil, and CLI command execution.
Sanitization: No evidence of input validation or sanitization for ingested natural language instructions before they influence task generation.

spec-kitty-agent