name learning-new-skills
description Performs fresh research on current AI/agents/skills standards and best practices, surveys public skill repositories for relevant patterns and ideas, and reports back what could improve the user's existing workflows. Use when invoked via /learn-new-skills, or when the user explicitly asks to check what's new in the AI tooling ecosystem.
disable-model-invocation true
allowed-tools WebSearch · Read

Learning New Skills

Stay current with the rapidly-moving AI/agents/skills ecosystem and surface ideas worth incorporating into the user's existing workflows (Yggdrasil, Odin Codin', and any others present).

Philosophy

Find ideas worth stealing, not skills worth installing. The goal is to report patterns, conventions, and gaps in the current setup — not to produce a download list. Bloat is the failure mode; keep it lean.

Process

  1. Check the learning_new_skills stamp in .meta/ledger.yaml (under skills:) to see when this skill was last run, so the research can focus on what's likely changed since then.

  2. Research current standards. Search for recent updates to:

    • agentskills.io specification and best practices
    • Anthropic's official Claude Code documentation
    • AGENTS.md conventions
    • Notable community write-ups from the last 30–60 days

  3. Survey published skills for patterns worth borrowing. Mine these sources for ideas to steal — not to install (see Philosophy). The big automated registries assume you install blind; we do the opposite, taking their best parts and making them ours.

    • skills.sh — the primary source. The install-ranked registry for the open skills ecosystem (Vercel's "npm for skills"). Its leaderboard surfaces what's actually catching on; survey the top skills and bundles to harvest structural patterns, conventions, and novel ideas. Treated as the primary index for now — currently the front-runner among the big skill directories.
    • Secondary — curated lists & reference repos:
      • VoltAgent/awesome-agent-skills
      • ComposioHQ/awesome-claude-skills
      • karanb192/awesome-claude-skills
      • anthropics/skills (official examples)
      • obra/superpowers (reference implementation of the brainstorm → plan → execute pipeline; check each run for new skills and patterns worth borrowing)
    • Other curated lists found during research.

    Look for: structural patterns, naming conventions, gotchas, novel uses of skills/subagents/commands, anything that addresses gaps in current workflows.

  4. Review existing workflows for fit. For each notable pattern found, ask: does any current Yggdrasil or workflow skill have a gap this would fill? Could an existing skill be sharpened by adopting this convention?

  5. Report findings. Produce a short summary covering:

    • What's genuinely new since last run (standards, tooling, conventions).
    • Patterns worth considering for adoption, with brief rationale.
    • Specific suggested improvements to existing skills, if any.
    • What was looked at and explicitly rejected as not-a-fit (so future runs don't re-evaluate the same things).

  6. Update the skills.learning_new_skills.last_run stamp in .meta/ledger.yaml with the current date. (The summary of what was found lives in the report from step 5 — the ledger holds only the date stamp.)

What this skill does NOT do

  • Download or install skills from external repos. That's a deliberate human decision after seeing the report.
  • Modify existing skills automatically. Suggest changes; let the user approve and apply them (or invoke a separate skill-editing flow).
  • Exhaustive surveys. Time-box the research; lean is better than thorough.

Output location

The report can be presented inline in the conversation, or saved to a dated file under working/research/YYYY-MM-DD-skills-scan.md if the user wants to keep a history. Ask which the user prefers if it's not clear.