arc-workflow-orchestrator（arc 工作流编排器）

Workflow Orchestrator

Chain skills into automated pipelines. Define a sequence of steps, and the orchestrator runs them in order with conditional logic, error handling, and optional audit logging.

Why This Exists

Agents run multiple skills but manually. Scan a skill, diff against the previous version, deploy if safe, log the result. That's 4 steps, 4 commands, and one missed step means a gap in your process. Workflows automate the sequence and ensure nothing gets skipped.

Commands

Run a workflow from a YAML file

python3 {baseDir}/scripts/orchestrator.py run --workflow workflow.yaml

Run a workflow from JSON

python3 {baseDir}/scripts/orchestrator.py run --workflow workflow.json

Dry run (show steps without executing)

python3 {baseDir}/scripts/orchestrator.py run --workflow workflow.yaml --dry-run

List available workflow templates

python3 {baseDir}/scripts/orchestrator.py templates

Validate a workflow file

python3 {baseDir}/scripts/orchestrator.py validate --workflow workflow.yaml

Workflow Format (YAML)

name: secure-deploy
description: Scan, diff, deploy, and audit a skill update
steps:
  - name: scan
    command: python3 ~/.openclaw/skills/skill-scanner/scripts/scanner.py scan --path {skill_path} --json
    on_fail: abort
    save_output: scan_result

  - name: diff
    command: python3 ~/.openclaw/skills/skill-differ/scripts/differ.py diff {skill_path} {previous_path}
    on_fail: warn

  - name: deploy
    command: python3 ~/.openclaw/skills/skill-gitops/scripts/gitops.py deploy {skill_path}
    condition: scan_result.verdict != "CRITICAL"
    on_fail: rollback

  - name: audit
    command: python3 ~/.openclaw/skills/compliance-audit/scripts/audit.py log --action "skill_deployed" --details '{"skill": "{skill_name}", "scan": "{scan_result.verdict}"}'
    on_fail: warn

Step Options

• name — Human-readable step name
• command — Shell command to execute (supports variable substitution)
• on_fail — What to do if the step fails: abort (stop workflow), warn (log and continue), rollback (undo previous steps), retry (retry up to 3 times)
• condition — Optional condition to check before running (references saved outputs)
• save_output — Save stdout to a named variable for use in later steps
• timeout — Max seconds to wait (default: 60)

Variable Substitution

Use {variable_name} in commands to reference:

• Workflow-level variables defined in the vars section
• Saved outputs from previous steps
• Environment variables with {env.VAR_NAME}

Built-in Templates

The orchestrator ships with these workflow templates:

1. secure-deploy — Scan → Diff → Deploy → Audit
2. daily-scan — Scan all installed skills, report findings
3. pre-install — Scan → Typosquat check → Install → Audit

Example: Secure Deploy Pipeline

name: secure-deploy
vars:
  skill_path: ~/.openclaw/skills/my-skill
  skill_name: my-skill
steps:
  - name: security-scan
    command: python3 ~/.openclaw/skills/skill-scanner/scripts/scanner.py scan --path {skill_path} --json
    save_output: scan
    on_fail: abort
  - name: deploy
    command: echo "Deploying {skill_name}..."
    condition: "CRITICAL not in scan"
    on_fail: abort
  - name: log
    command: python3 ~/.openclaw/skills/compliance-audit/scripts/audit.py log --action workflow_complete --details '{"workflow": "secure-deploy", "skill": "{skill_name}"}'

Tips

• Start with --dry-run to verify your workflow before executing
• Use on_fail: abort for security-critical steps
• Chain with the compliance audit skill for full traceability
• Keep workflows in version control for reproducibility