Create an OpenClaw Skill

Good Skills are narrow, explicit, and easy to audit. Use the smallest instruction set that changes behavior reliably.

Create the directory

skills/
└── release-notes/
    └── SKILL.md

Add required frontmatter

---
name: release-notes
description: Create release notes from a repository diff when asked for a changelog or release summary.
---

Use a stable lowercase name. The description should say both what the Skill does and when it should be selected.

Write operational instructions

Prefer a short sequence with observable checks:

1. Identify the comparison range.
2. Read the diff and affected documentation.
3. Separate user-visible changes from internal maintenance.
4. Flag anything not proven by the diff.
5. Return the requested format.

Add safety boundaries where the task may mutate state. State what needs confirmation and what sources are authoritative.

Avoid common mistakes

• Do not put plugin JavaScript boilerplate in a plain Skill.
• Do not include secrets or real tokens in examples.
• Do not download and execute remote scripts without a clear trust check.
• Do not claim broad capability in the description when the instructions cover one narrow workflow.

Test the trigger and the output

Start a new session, ask for a task that should trigger the Skill, and also test a nearby task that should not. Verify that the agent reads the file, follows the sequence, and respects boundaries.

Use the static SKILL.md Checker as a preflight, then test with OpenClaw itself. A static checker can flag patterns; it cannot prove runtime safety.

Expected result

The Skill loads under its intended name, triggers for the target task, stays inactive for a nearby unrelated task, and completes the workflow without exceeding its documented permissions.