The Agent Skills Directory

[SAFE]: The skill implements a robust 'human-led curation' model. It specifically states that 'The human decides' and 'Only write what the human approved,' which ensures all generated outputs and file system changes (to .oh/metis/ or .oh/guardrails/) are reviewed by the user, preventing automated execution of potentially untrusted data.- [SAFE]: The ingestion of previous session data via 'oh_search_context' represents a potential indirect prompt injection surface; however, the risk is mitigated by the lack of automated tool execution and the requirement for manual verification of clustered themes and proposals. Ingestion point: oh_search_context (SKILL.md). Boundary markers: None. Capability inventory: Local file writes to .oh/ directory. Sanitization: Human-in-the-loop review process.- [NO_CODE]: The skill consists entirely of instructional markdown and does not include any executable scripts, binaries, or external package dependencies.

distill