The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill mentions an optional CLI package, @osovv/grace-cli, which is provided by the same author ('osovv') for linting and status checks. This is a standard vendor-provided utility for the methodology.
[COMMAND_EXECUTION]: The documentation references various CLI commands (e.g., grace lint, grace status) and agent-specific triggers (e.g., $grace-plan, $grace-execute). These are intended for project management and architectural verification as part of the GRACE methodology.
[INDIRECT_PROMPT_INJECTION]: The methodology is designed to process external project files, including XML artifacts and source code with semantic markup, to maintain context. This creates a surface where the agent ingests untrusted data from the project being worked on.
Ingestion points: docs/knowledge-graph.xml, docs/requirements.xml, docs/development-plan.xml, docs/verification-plan.xml, and source code files (src/**/*).
Boundary markers: The system uses explicit delimiters like START_MODULE_CONTRACT / END_MODULE_CONTRACT and START_BLOCK_NAME / END_BLOCK_NAME to scope data.
Capability inventory: The skill allows for shell command execution via the grace CLI and agent-led code generation/modification.
Sanitization: The methodology focuses on structural integrity but does not explicitly document sanitization of external text content against embedded instructions.

grace-explainer