The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to perform availability checks on local orchestration tools (e.g., maestro ralph skills) and to execute CLI commands defined within workflow templates. These operations are limited to the local environment and the tool's intended management scope.
[PROMPT_INJECTION]: The skill processes untrusted user input via the $ARGUMENTS variable to classify intent and select task chains, which constitutes an indirect prompt injection surface.
Ingestion points: User intent is received through the $ARGUMENTS parameter in SKILL.md.
Boundary markers: The skill generates boundary_contract and execution_criteria fields in the status.json session file during the A_DECOMPOSE_TASKS phase to constrain agent actions.
Capability inventory: The skill has access to Bash for command execution, Write and Edit for session state persistence, and spawn_agent for delegating tasks to sub-agents.
Sanitization: Input is parsed through a semantic classification engine (A_CLASSIFY_INTENT) and validated against a pre-verified list of project and global skills before any execution occurs.

maestro