Skip to main content

Skill Harmonization

Also available in workflows

This specialist runs in chat and as a node you can add to a workflow for repeatable, scheduled runs. See the Agent Catalog for how the two relate.

Overview

Skill libraries drift over time. Different teams add their own wording, near-identical skills pile up, and the list grows faster than anyone can clean it. Skill Harmonization reviews a set of skills and recommends what to do with each one.

Ellie sorts your skills into three groups: those to consolidate (near-duplicates that should be merged under one clear name), those to maintain as they are, and those to discard. Every recommendation comes with a reason, so you see the thinking rather than just a verdict.

The analysis is deliberately conservative. Ellie works only from the skills you provide, never inventing skills or relationships, and only merges or drops a skill when the evidence is strong. Where the data is missing or ambiguous, it flags the uncertainty instead of guessing.

What you'll need

  • A list of your current skills — a spreadsheet, a framework export, or the skills you produced with Skill Extraction. Attach it to the conversation or add it as context.
  • The more detail each skill carries (category, description, level), the more reliable the merge decisions.
  • Optionally, a target strength: weak (about a 20% reduction), medium (about 40%, the default), or strong (about 60%).

How to ask

Point Ellie at your skills list and say what you want cleaned up.

Consolidate an existing library:

Here is our current skills library (attached). Consolidate the overlapping
skills and tell me which to merge, keep, or drop.

Ask for a specific reduction strength:

Review the attached skills list and give me a strong consolidation —
aim for roughly a 60% reduction in the number of skills.

Clean up duplication across teams:

We have too many near-duplicate skills across departments. Cluster the
attached inventory and recommend a cleaner, deduplicated set.

What you get

A Skill Harmonization report in four parts:

  • Skills to Consolidate — groups of near-duplicate skills, the skills in each group, a suggested merged name, and the reason for merging.
  • Skills to Maintain — skills to keep as they are, with their category and why they stay.
  • Skills to Discard — skills recommended for removal, each with a reason.
  • Verdict and Summary — a short written assessment of the overall reduction and its impact.

Tips

  • Provide as complete a list as you can. Recommendations are made only from what you give it, so anything missing simply won't be considered.
  • Set the strength to match your goal — a light tidy-up (weak) or an aggressive consolidation (strong).
  • Include categories and descriptions where you have them; richer detail makes merge decisions more dependable.
  • Treat the output as a recommendation. Review merged names so they fit your organization's terminology before you apply them.
  • Anything ambiguous is flagged rather than resolved automatically, so watch for those notes.