harness-skill-author
Skill1 starsChanged 20 days agoApache-2.0View on GitHub
Use when creating, reviewing, or improving an Antigravity skill (SKILL.md), or when a repeated workflow should become a reusable skill. Covers frontmatter, trigger descriptions, progressive disclosure, verification, and safe attribution.
In this repository7 other files it gives its agents.
- rules/AGENTS.mdAGENTS.md
- harness-controlSkill
- harness-handoffSkill
- harness-recallSkill
- harness-reflectorSkill
- systematic-debuggingSkill
- verification-before-completionSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 6 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 24
Do not write one for a single command, a one-off incident, or a workflow that never succeeded.
- Very specific: line 53
Keep the main file short. Move long references, scripts, and examples into references/, scripts/, or assets/ and link them.
- Distinctive: line 63
Procedure: use commands that can be copied and run.
- Distinctive: line 64
Prefer the project's own scripts over ad-hoc shell.
- Distinctive: line 68
No secrets: no tokens, keys, cookies, private URLs, or customer data.
- Distinctive: line 69
Remove paths that identify a person.
- Distinctive: line 70
No transcript noise: no timestamps, UUIDs, bare exit codes, or truncated output.
- Very specific: line 72
Attribution: if you copy or adapt someone else's text, keep their license notice and a source link in the skill and in NOTICE.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Harness² Skill Author
A skill is a short procedure that an agent can reuse. It is not a changelog, a tutorial, or a saved transcript.
When a skill is worth writing
Write one when:
- the same workflow will come back in other sessions or projects;
- the steps are stable enough to repeat;
- there is a real way to check the result;
- another agent would otherwise rediscover the setup, flags, or traps.
Do not write one for a single command, a one-off incident, or a workflow that never succeeded.
Shape
---
name: kebab-case-name
description: >-
Use when ... Name the symptoms, tools, and task types that should trigger
this skill. Keep it under 1024 characters.
---
# Title
## When to Use
- concrete triggers
## Procedure
1. command or action, with the expected result
2. ...
## Verification
- how to prove it worked
## Pitfalls
- error → verified fixKeep the main file short. Move long references, scripts, and examples into
references/, scripts/, or assets/ and link them. That is progressive
disclosure: the description is always visible, and the body loads only when the
skill is chosen.
Writing rules
- Name: lowercase kebab-case, identical to the folder name.
- Description: explain when to use it, not only what it is. Include likely keywords and error strings.
- Procedure: use commands that can be copied and run. Mark placeholders. Prefer the project's own scripts over ad-hoc shell.
- Verification: require something observable, such as a test, build, health check, exit code, or file.
- Pitfalls: include only fixes that were actually verified.
- No secrets: no tokens, keys, cookies, private URLs, or customer data. Remove paths that identify a person.
- No transcript noise: no timestamps, UUIDs, bare exit codes, or truncated output.
- Attribution: if you copy or adapt someone else's text, keep their license
notice and a source link in the skill and in
NOTICE.
Review checklist
- Frontmatter parses; the name is kebab-case; the description names triggers.
- At least two distinct steps; three for a generated skill.
- Verification is explicit.
- Check for duplicates with
bin/harness skills list. bin/harness skills validatepasses.- No credentials or private data.
- The file is well below the 30 KB hard limit.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

