This skill treats SKILL.md, AGENTS.md, CLAUDE.md, and small local guidance files as variants of the same pattern: a routing file. A routing file is a compact entrypoint that states ownership, identifies intent, and points to the smallest useful endpoint.
<mental_model> Every file has one job:
- Routing files answer "where should this go next?"
- Workflows answer "what steps should be followed?"
- References answer "what should be known?"
- Templates answer "what shape should output take?"
- Scripts answer "what repeatable check or action should run?"
If a file starts doing more than one job, split it or demote the extra detail into a more specific endpoint. </mental_model>
<quick_start> When invoked:
- Identify the user's intent.
- Decide whether the task is routing, authoring, auditing, templating, or validation.
- Load one matching endpoint from
<routing>. - Follow that endpoint and stop loading context once the task is clear.
- Create or update the smallest file that owns the needed behavior.
Ask a question only when two routes would produce meaningfully different files. </quick_start>
| intent | endpoint |
|---|---|
| Decide where guidance belongs | workflows/route-information.md |
| Create or rewrite a routing file | workflows/write-routing-file.md |
| Create a workflow endpoint | workflows/write-workflow.md |
| Create a reference endpoint | workflows/write-reference.md |
| Create or revise templates | workflows/write-template.md |
| Add helper scripts or checks | workflows/write-script.md |
| Audit framework structure | workflows/audit-framework.md |
| Validate files against schemas | workflows/validate-schema.md |
| Browse schema definitions | schemas/README.md |
| Create tests for a hook | workflows/write-hook-tests.md |
| Understand @-linking mechanics | references/at-linking.md |
| Apply claude.md routing rules | references/claude-md-routing-rules.md |
| Understand the core model | references/routing-framework.md |
| Choose XML tags | references/xml-tags.md |
| Use naming and ownership rules | references/naming-and-ownership.md |
<routing_file_variants>
| variant | scope |
|---|---|
SKILL.md |
Reusable capability or package entrypoint |
AGENTS.md |
Repository or working-context entrypoint |
CLAUDE.md |
Claude-specific project entrypoint |
simple-skill |
Minimal routing file for one small capability |
| Nested router | Subdomain entrypoint inside a larger framework |
These are not separate concepts. They are routing files with different scope and runtime conventions. </routing_file_variants>
<framework_rules>
- Keep routing files small.
- Route by intent before file type.
- Put durable concepts in references.
- Put ordered procedures in workflows.
- Put reusable output shapes in templates.
- Put repeatable inspection or execution in scripts.
- Keep endpoint references one level deep from the router that names them.
- Prefer one focused endpoint over one mixed-purpose file.
- Use semantic XML sections in agent-facing bodies.
- Keep markdown headings out of XML-structured bodies.
- Preserve existing runtime conventions when adapting to
AGENTS.md,CLAUDE.md, or another host format.
</framework_rules>
<success_criteria> The framework is healthy when:
- A new task reaches the right endpoint quickly.
- The root router remains readable in one pass.
- Similar entrypoints share one routing-file model.
- Workflows, references, templates, and scripts have predictable homes.
- Helper scripts can inspect the framework without becoming required for comprehension.
- Old or competing routing concepts are absorbed into the unified model. </success_criteria>