| name | adapting-skills |
| description | Use when evaluating, adopting, absorbing, or adapting specified existing agent skills into a local workflow-compatible skill set without skill sprawl. |
| license | MIT |
| metadata | {"author":"h3y6e","version":"2026.5.5"} |
Adapting Skills
Adapt specified source skill(s) into a smaller, local, workflow-compatible skill set. The default is to reduce skill count.
REQUIRED SUB-SKILL: Use authoring-skills before creating or editing any SKILL.md. This skill chooses adopt/adapt/absorb/reject; authoring-skills controls drafting, RED scenarios, description quality, and validation evidence. For external candidates, use references/adoption-rubric.md.
Inputs
Read only needed inputs:
- source skill(s): local paths, installed skills, URLs, or pasted text
- target workflow constraints and existing project instructions
- already adopted skills and repo-local skills
- named validation expectations
Decisions
Classify each source skill:
| Decision | Use When |
|---|
adopt as-is | It is standalone, compatible, and does not conflict with target workflow. |
adapt | The core workflow is valuable but source conventions, tool syntax, or scope must change. |
absorb | Only a small idea is useful; add it to an existing skill instead of creating a new one. |
reject | It is redundant, one-off, too broad, too tool-specific, or not worth the context cost. |
Prefer adopt as-is or absorb over adapt. Create a new skill only when it has standalone recurring value.
Adaptation Rules
- Preserve the source skill's useful behavior, not its incidental naming, directory layout, or tool assumptions.
- Remove setup workflows, private conventions, and lifecycle state machines unless the target workflow already uses them.
- Merge overlapping triggers instead of keeping competing skills.
- Keep resulting
SKILL.md files under 500 words when practical.
- Description must say when to use the skill, not summarize its workflow.
- Add output contracts, stop conditions, and validation scenarios for behavior-changing skills.
- Do not claim empirical validation unless a fresh evaluator actually ran the scenarios.
Process
- Read the source skill(s) and identify the recurring behavior they protect.
- Compare against already adopted and repo-local skills.
- Produce a decision table before editing.
- Apply the smallest change: adopt, absorb, adapt, or reject.
- Update README or catalogs only for skills that remain.
- Record validation status:
untested, structurally reviewed, or scenario-tested.
Output
## Decision Table
| Source skill | Decision | Reason | Destination |
| --- | --- | --- | --- |
## Changes
- <skill or file>: <what changed and why>
## Rejected
- <source skill>: <reason>
## Adoption Evidence
- Source snapshot:
- Rubric axes checked:
- Redundancy/compatibility evidence:
- License/provenance:
## Validation
- Structural checks:
- Scenario status:
- Empirical status:
Common Mistakes
- Forking a whole skill when one paragraph should be absorbed.
- Keeping source-specific setup because it was in the original.
- Creating a router skill that duplicates descriptions.
- Saying a skill is validated after only static review.