wesleymatosdev/learning-routing

Classify learnings into memory, skills, or project records.

¿Qué es learning-routing?

learning-routing is a Claude Code agent skill that classify learnings into memory, skills, or project records.

Compatible conClaude Code~Codex CLI~Cursor
npx skills add https://github.com/wesleymatosdev/skills/tree/main/workflow/learning-routing

Installed? Explore more Productividad y colaboración skills: steipete/gemini, steipete/gh-issues, steipete/skill-creator · View all 6 →

Preguntar en tu IA favorita

Abre un nuevo chat con esta habilidad de agente ya precargada.

Documentación

Learning Routing

Use a deliberate routing decision whenever feedback, an incident, or a solution could become durable knowledge. Keep global context compact; preserve detailed evidence where it remains useful.

When to Use

  • A user correction reveals a repeatable standard or preference.
  • A bug fix or delivery taught a reusable lesson.
  • A project has accumulated memories that may be candidates for Hermes memory or skills.
  • You are about to write durable memory or create a skill.

Do not use this to migrate an entire project corpus wholesale.

Procedure

  1. Separate the durable rule from the incident.

    • Record what happened, exact system details, and evidence in the project record.
    • Extract only the broadly reusable rule for further classification.
    • Completion: the proposed global item stands alone without repo-specific identifiers.
  2. Decide scope.

    • Use user memory for a compact, stable fact about the user or an enduring preference.
    • Use a skill for a reusable, multi-step procedure that needs on-demand instructions.
    • Use a project record for domain rules, infrastructure details, team conventions, credentials locations, current state, or a one-off incident.
    • Completion: choose one primary home; do not duplicate detailed content across all three.
  3. Decide delivery.

    • Keep always-loaded memory to a short declarative fact.
    • Put conditional steps, examples, commands, and pitfalls in a skill.
    • Keep detailed reasoning, rejected alternatives, and evidence with the project.
    • Completion: the item will load only when its detail is useful.
  4. Check portability and sensitivity.

    • Do not promote employer-, repository-, or account-specific details merely because they appear across related repositories.
    • Do not promote personal or sensitive information without a clear task benefit and user approval.
    • Completion: the destination does not disclose information beyond its intended scope.
  5. Preserve provenance.

    • When promoting a rule, leave the project source intact or archive it rather than deleting it.
    • State what was promoted and what intentionally remained project-scoped.
    • Completion: a future agent can trace the global rule back to its evidence.

Pitfalls

  • A cross-repository fact is not automatically portable to unrelated projects.
  • A large global memory or a giant skill reduces relevance and adherence.
  • A passing test or a confident agent summary is evidence only when it proves the claimed behavior.

Verification

Before finalizing a promotion, verify that each reviewed item is accounted for as memory, skill, project record, or discarded as stale. Report the count and the reason for each promoted item.

Individual skills in this repo

This repo contains 2 individual skills — each has its own dedicated page.

Skills relacionados