Customizing skills

PreviousNext

Edit, fork, disable, or write your own skills.

Edit a managed skill

You can edit files in .agents/skills/<skill>/ directly. AgileFlow notices the change (list shows modified, check warns) and will not overwrite it during updates. But a modified managed skill is stuck between two owners: every future update asks you to fork, reset, or skip.

Fork to own it

If you want to keep your version, fork it:

agileflow fork filing-pr

The skill becomes locally owned: agileflow.yaml points at .agents/skills/filing-pr with provenance.forkedFrom, updates never touch it, and agileflow diff filing-pr --upstream shows what upstream changed since. See fork and diff.

Forking is a feature, not a failure. If many people fork a skill to delete half of it, the official skill is too prescriptive, and that feedback goes back into it.

Change how a skill activates

agileflow configure skill filing-pr --activation manual

AgileFlow writes each provider's manual-only flag into the installed skill. For forks, edit the flags yourself (see Providers).

Disable without removing

agileflow configure skill reviewing-changes --disable

Write your own

Create .agents/skills/my-release-flow/SKILL.md:

---
name: my-release-flow
description: Cut a release of this service. Use when the user asks to release, tag, or publish a version.
---
 
1. Update CHANGELOG.md from merged PRs.
2. Run `make release-check`.
3. Tag and push.
 
## Done when
 
The tag is pushed and the release workflow is green.

Every provider sees it immediately; Claude Code sees it once it is linked. AgileFlow leaves unmanaged skills alone. To have AgileFlow track it (for Claude Code links and check), register it in place:

agileflow add ./.agents/skills/my-release-flow

To share a skill across repositories, keep it in its own git repository and agileflow add git+https://....

Guidance that pays off: the description is routing logic, so say what the skill does and when to use it. Keep the body short and write down only what the model cannot infer: repository-specific steps, constraints, and failure modes you have actually seen.