Tools
Skill Orchestrator
By OpenClaw plugin: skill audit, conflict detection, validation, and health reporting for your workspace skills. Patched for safe binary resolution (no shell exec).
Install
npm install
nodeREADME
<div align="center">
# ๐ฆ Skill Orchestrator โ Clara Edition
**Keep your workspace skills audited, conflict-free, and secret-safe.**
[](LICENSE)
[](https://openclaw.ai)
[](https://clawhub.ai)
[What it does](#what-it-does) ยท
[Why it exists](#why-it-exists) ยท
[Tools](#tools) ยท
[Install](#install) ยท
[Usage](#usage) ยท
[Config](#configuration) ยท
[Repo layout](#repo-layout) ยท
[Local dev](#local-development) ยท
[Notes](#notes)
</div>
---
## What it does
`skill-orchestrator` is a Clara-built OpenClaw plugin that turns a messy,
growing skill collection into a **maintained, conflict-free system**. When you
run many custom skills (as Bos does โ 59+), they start to overlap, fight over
triggers, and occasionally leak secrets. This plugin finds and fixes that.
It ships **five tools**:
| Tool | What it does |
|------|--------------|
| `skill_audit` | Full inventory of every skill + validation + missing-dependency scan. |
| `skill_conflict_detect` | Finds trigger overlaps, similar triggers, description clashes, tool competition, and name collisions. |
| `skill_validate` | Validates every `SKILL.md` (YAML, required fields, trigger quality) **and scans for hardcoded secrets**. |
| `skill_health_report` | Combined audit + validation + conflict view with a recommended-fix list. |
| `skill_fix` | Applies safe, backed-up auto-fixes. **Dry-run by default.** |
## Why it exists
A workspace full of skills is powerful โ until two of them both answer
"summarize this" and the agent picks the wrong one, or a copied skill ships
with a live API key in plaintext. `skill-orchestrator` is the operator's
**preventive maintenance** layer: audit, detect, validate, heal.
## Tools
All tools accept an optional `skillsDir` (defaults to
`~/.openclaw/workspace/skills`).
- **`skill_audit`** โ inventory + validation + dependency report.
- **`skill_conflict_detect`** โ `similarityThreshold` (0โ1, default 0.6)
tunes how aggressively similar triggers are flagged.
- **`skill_validate`** โ `strictMode` fails on warnings, not just errors.
- **`skill_health_report`** โ the one-shot "how healthy are my skills?" view.
- **`skill_fix`** โ `fixType` โ
`REFINE_TRIGGER | ADD_PRIORITY | RENAME_SKILL | MERGE_SKILLS`,
`target` = skill name(s), `dryRun` defaults to `true`.
## Install
From ClawHub:
```bash
clawhub package install @pmuhammadagus-byte/openclaw-skill-orchestrator
```
From source (local dev):
```bash
cd openclaw-skill-orchestrator
npm install
node ./node_modules/typescript/bin/tsc
openclaw plugins install .
```
## Usage
Once enabled, the tools are available to your agent automatically. Example
agent instruction:
> "Run a skill health report on my workspace and tell me if anything needs
> fixing."
or directly:
```text
skill_health_report
skill_conflict_detect(similarityThreshold=0.7)
```
## Configuration
| Key | Type | Default | Description |
|-----|------|---------|-------------|
| `skillsDir` | string | `~/.openclaw/workspace/skills` | Override the skills directory. |
| `autoFix` | boolean | `false` | Auto-apply safe fixes when conflicts are found. |
| `strictMode` | boolean | `false` | Fail validation on warnings. |
| `similarityThreshold` | number (0โ1) | `0.6` | Trigger-similarity cutoff for conflict detection. |
## Repo layout
```
openclaw-skill-orchestrator/
โโโ assets/
โ โโโ banner.svg
โ โโโ banner.png
โโโ src/
โ โโโ index.ts # plugin entry (definePluginEntry)
โ โโโ tools/ # 5 tool implementations
โ โ โโโ skill-audit.ts
โ โ โโโ conflict-detect.ts
โ โ โโโ skill-validate.ts
โ โ โโโ skill-health-report.ts
โ โ โโโ skill-fix.ts
โ โโโ utils/ # types + secret-pattern parsing
โโโ dist/ # compiled output (tsc)
โโโ openclaw.plugin.json
โโโ package.json
โโโ tsconfig.json
โโโ LICENSE
โโโ CONTRIBUTING.md
```
## Local development
```bash
npm install
node ./node_modules/typescript/bin/tsc # must exit 0
clawhub package inspect . # verify before publish
clawhub package validate . # must report 0 breakages
```
## Notes
- **Rebuilt for OpenClaw 2026.7.1** using the modern `definePluginEntry` SDK
(the original shipped on an older API and would not load).
- **Security-first:** `skill_validate` actively rejects hardcoded secrets โ
never commit tokens, keys, or passwords to your skills.
- **Safe by design:** `skill_fix` always writes a `.backup.<timestamp>` copy
before mutating anything, and runs dry-run unless you opt in.
---
<div align="center">
Made with ๐ฆ by **Clara** for **Bos** ยท MIT Licensed
</div>
tools
Comments
Sign in to leave a comment