Documentation
CI/CD Pipeline API & Skill Testing Reference
Technical reference for the ClawProd pipeline API, skill testing framework, and security scanning configuration.
Pipeline API
The ClawProd pipeline runs automatically on every git push via a GitHub Action. In this local preview, the dashboard uses the pipeline endpoints below.
Trigger a Build
Run Evidence Gate
Get Build Status
List Builds
Pipeline Stages
Every build runs five stages in order. A failure at the evidence gate blocks publish.
- Lint — Validates skill manifest, metadata fields, and file structure against the OpenClaw skill spec.
- Test — Runs your test suite in an isolated sandbox. If no custom tests exist, runs the default test suite (manifest validation, permission checks, basic execution).
- Build — Packages the workspace and prepares the release artifact for scanning.
- Evidence Gate — Runs the shared
@claw/evidencescanner against OpenClaw artifacts and emits JSON control evidence. - Publish — On pass, bumps version, generates changelog, and publishes to ClawHub. Only runs if
publish: true.
Skill Testing Framework
ClawProd includes a built-in testing framework for OpenClaw skills. Tests run in a sandboxed environment with a real OpenClaw instance.
Test File Structure
Default Test Suite
Skills without custom tests still get checked against the default suite:
- Manifest schema validation (required fields, correct types)
- Permission scope verification (no unnecessary permissions)
- Basic execution test (skill loads and responds without error)
- Dependency version check (no known-vulnerable packages)
Security Scanning Reference
The security scanner runs five checks on every build:
- CVE scan — Checks all dependencies against the National Vulnerability Database.
- Permission analysis — Flags skills requesting more permissions than their functionality requires.
- Data exfiltration detection — Identifies patterns like unauthorized network calls, file system reads outside scope, or credential harvesting.
- Prompt injection vectors — Scans for common injection patterns in skill prompts and input handling.
- Custom policies — Enterprise users can define organization-specific security rules.
GitHub Action Setup
The first release gate is the shared OpenClaw evidence scanner. It reads an AGENTS.md, SOUL.md, or skill config and emits a JSON artifact that CI can fail on.