The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data which could contain malicious instructions intended to influence the rule distillation process.
Ingestion points: Processes "skills, prompts, traces, reviews, failures, or docs" as defined in the source corpus section of SKILL.md.
Boundary markers: Implements structured artifacts such as principle_candidate_set/v1 and review_queue/v1 to isolate and categorize candidate rules before they are processed further.
Capability inventory: The skill has access to skill_manage for modifying files and utilizes the omh CLI for runtime logging and status recording.
Sanitization: Employs a mandatory human-in-the-loop review process as a primary defense; the instructions explicitly forbid silent mutation of system files and require explicit approval recorded in the review_queue/v1.
[COMMAND_EXECUTION]: The skill utilizes a platform-specific command-line tool for operational tracking and telemetry.
Evidence: Execution of omh runtime record --skill rules-distill --harness rules-distill --status started to signal the start of the workflow execution.

omh-rules-distill