The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local command-line tools grace lint and grace status to validate the environment and project state.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from external XML files to generate an implementation plan containing executable commands.
Ingestion points: Data is read from .grace/changes/active/C-CHANGE-ID/spec.xml and design-context.xml in the local file system.
Boundary markers: The skill relies on XML tags for structure but lacks explicit delimiters or instructions to the agent to disregard potential instructions embedded within the text content of the input fields.
Capability inventory: The skill invokes the grace CLI and generates a plan.xml file that defines shell commands for task verification.
Sanitization: The instructions do not define any sanitization, filtering, or validation for the content extracted from input files before it is used to populate the verification commands in the output plan.

grace-plan