The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local initialization script scripts/gcc_init.sh to set up the .GCC/ directory structure. This script is part of the skill's own package and performs standard file-system operations (mkdir, cat) without external dependencies.
[INDIRECT_PROMPT_INJECTION]: The skill maintains project state and history in Markdown and YAML files, which are read during context recovery. This creates a surface where external content (if committed to the history) could influence the agent.
Ingestion points: The agent reads main.md, metadata.yaml, commit.md, log.md, and branch summary.md files to reconstruct context.
Boundary markers: No specific boundary markers or 'ignore' instructions are defined for the content read from these files.
Capability inventory: The skill allows file creation, directory management, and local script execution (gcc_init.sh).
Sanitization: No explicit sanitization or validation of the retrieved context is performed beyond the agent's internal reasoning.
[EXTERNAL_DOWNLOADS]: The skill documentation mentions installation via npx skills add, which is the standard distribution method for this environment. No unexpected or untrusted remote code execution patterns were identified.

gcc