The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes contents from project-specific files such as M###-CONTEXT.md, S##-CONTEXT.md, and .gsd/DECISIONS.md. While these ingestion points represent an attack surface where untrusted data could influence agent recommendations, the skill follows a strict human-in-the-loop process (one question at a time) and requires explicit user confirmation for all material outputs and actions.
[COMMAND_EXECUTION]: The skill uses a platform-specific sub-agent spawning mechanism (Agent(subagent_type=Explore)) to map the codebase. This is a legitimate use of agent capabilities for context gathering within the local environment.
[EXTERNAL_DOWNLOADS]: The skill references the mcp__github__issue_write tool for creating issues. The instructions explicitly state that this tool must only be used with human confirmation, adhering to safe practices for outward-facing actions.

grill-me