This repository is the curated source of truth for mature, reusable agent skills. Drafting, debate, and exploratory evaluation happen outside this repository; only reviewed and validated skills graduate here.
- context-scan — Explicit, bounded diagnosis of task-context interference with threshold-controlled remediation.
Each skill lives under skills/<skill-name>/ and contains a required SKILL.md plus only the supporting resources needed by its workflow.
Validate a skill with the current Codex skill validator before publishing or revising it. Review validation should cover trigger precision, least authority, bounded context use, bounded execution, recursion safety, and explicit stopping conditions.