What problem does it solve? Deciding where new agent content belongs โ a rule, a SKILL.md, a reference file, or an asset โ is ambiguous, and wrong choices bloat context budgets or bury information behind chained reads. This Skill provides container tiers, link-depth rules, and an aggregation decision procedure for structuring agent libraries under .cursor/ or a plugin stack. ## Core Features & Use Cases - Container tier guidance: Maps content types (always-on rules, glob rules, SKILL.md, references, assets, AGENT.md) to their context cost so each piece lands in the cheapest tier that fits. - Aggregation decision procedure: A four-step test run before creating any new folder, preferring deepening an existing skill with references over widening the tree with sibling skills. - Drift signals and restructuring: A table of warning signs (SKILL.md past 500 lines, chained references, underpopulated subfolders) with concrete corrective actions, illustrated by worked examples in references/worked-examples.md. - Use Case: When adding pytest guidance to an agent library, apply the nesting test to decide it belongs as references/testing.md under the Python language skill rather than a new quality/testing sibling. ## Quick Start Ask the agent to apply the agent-hierarchy rules to decide whether a new piece of guidance should become a rule, a reference, or a new skill.
