The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/detect_gpc.mjs script uses execSync to run CLI commands for environment detection. * Evidence: Commands such as gpc --version, npx gpc --version, and gpc auth status are executed to gather status information. * Context: These operations are hardcoded and limited to discovery of the local GPC installation and configuration.
[INDIRECT_PROMPT_INJECTION]: The skill architecture facilitates plugins that process command and API data, which is an inherent surface for indirect prompt injection. * Ingestion points: Plugin hooks defined in references/hooks-reference.md receive data from command results and HTTP responses. * Boundary markers: The framework includes automated redaction of sensitive credentials and tokens in command metadata. * Capability inventory: The plugin system includes a manifest-based permission model (read:auth, api:read, etc.) as documented in references/permissions-system.md. * Sanitization: Documentation in references/hooks-reference.md instructs developers to use the sensitive property for redacting custom arguments and options.

gpc-plugin-development