The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to invoke a command-line utility called grace to execute change plans. It specifically enables the execution of commands defined as 'MustPassCommand' within these plans. This execution is the core purpose of the skill and is governed by strict operational rules.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external files (spec.xml and plan.xml). This ingestion creates a surface where external content could influence the agent's actions. However, the skill provides instructions to only use plans that have undergone an external approval and identity-matching process. 1. Ingestion points: Configuration and plan details are read from spec.xml and plan.xml files as described in the preflight section. 2. Boundary markers: The skill relies on an external 'approved' status and 'identity-matched' checks to ensure the integrity of the input files. 3. Capability inventory: The agent performs shell command execution using the grace tool during baseline, target, and final validation phases. 4. Sanitization: The skill does not explicitly define internal sanitization for the commands found in the plans, relying instead on the tool's structured plan execution logic.

grace-execute