Local skill builder · Independent guidance

Create an OpenClaw SKILL.md with safer triggers

A strong skill states exactly when it should and should not run, declares every tool, command, account, variable, and service it depends on, and separates instructions from risky external actions. The builder produces a local draft and test checklist; it does not install or verify the skill.

Current snapshotBrowser-only toolUpdated 2026-08-24Official docs main snapshot · OpenClaw 2026.8.1

Draft + lint locally

SKILL.md builder

Does not install or publish

Skill nameTrigger descriptionPositive triggers, comma separatedNegative and safety boundariesDependencies, semicolon separatedInstructions

DRAFT

Ready for isolated testing

Lint passed

---
name: openclaw-health-review
description: "Review OpenClaw health output when the user asks to inspect doctor or gateway status; do not run fixes or change configuration."
---

# openclaw-health-review

## Activate when

- review doctor output
- explain gateway health
- summarize findings

## Do not activate / safety boundaries

- do not install software
- do not run --fix
- do not expose secrets

## Dependencies

- openclaw CLI
- official OpenClaw docs

## Instructions

1. Ask for the smallest redacted JSON report.
2. Separate observations from actions.
3. Cite the official check or CLI page.
4. Stop before any state-changing command.

## Test cases

- Positive: user provides redacted doctor JSON and asks for an explanation.
- Negative: user asks to install or update OpenClaw; this skill must not activate.
- Failure: input is unknown; state the limitation and request official structured output.
- Safety: input includes a token; mask it and do not repeat or store it.

Generated means ready to review—not safe, installed, verified, or accepted by ClawHub.

Minimum skill contract

A valid, narrow name and a description containing positive triggers.
Negative boundaries that prevent adjacent or high-risk activation.
Declared commands, services, accounts, environment variables, and resources.
Explicit approval, target resolution, rollback, and credential handling for external writes.
Positive, negative, failure, and safety test cases.

Test before applying

Place the draft in an isolated workspace, inspect the final package tree, run the OpenClaw skills checks available to your version, and exercise negative triggers as carefully as the happy path. A generated file is only ready for testing, never automatically trusted.

Common questions

Before you act

Does the builder publish to ClawHub?

No. It produces a local Markdown draft and checklist only.

Can I paste an existing skill?

Yes. Import remains local; unknown text is preserved while the builder reports supported metadata and safety observations.

Primary sources

Verify against the owner.

Creating OpenClaw skillsofficial docs · verified 2026-08-24

OpenClaw skills systemofficial docs · verified 2026-08-24

OpenClaw skills CLIofficial docs · verified 2026-08-24

Content snapshot follows official docs main at the recorded commit; the current package metadata reported 2026.8.1 when verified. Your installed release and live CLI schema remain authoritative.