Overview
Agentic Workflow Template Builder generates production-ready workflow templates for autonomous agent pipelines before a single line of automation code runs. Choose from five workflow types — admin, casework, research, support, or custom — or supply your own step list, and get back three output files: a machine-readable YAML workflow contract, an operator playbook in Markdown, and a validation test plan.
Every template includes roles, human approval gates, evidence requirements, retry policy, stop conditions, privacy boundaries, and handoff outputs — the complete operating skeleton that makes agent workflows auditable, reversible, and safe to run unattended.
Key Features
- Five built-in workflow types: admin, casework, research, support, custom
- Human approval gates auto-applied to decision, execution, and publish steps
- Evidence requirements, retry policy, and stop conditions per step
- Custom step names via comma-separated
--stepslist - Three output files: YAML workflow contract, Markdown playbook, test plan
- YAML injection prevention: control character rejection and safe scalar quoting
- No network access, no subprocess calls, no API keys required
Use Cases
- Government and public sector: casework and admin workflows with human approval gates before autonomous deployment
- Operations builders: intake→triage→decision→handoff pipelines for support queues and case management
- AI/agent developers: rapid workflow skeleton prototyping for multi-step agent sequences
- Compliance teams: workflow contracts documenting owner, evidence, and privacy boundaries for audit approval
Example Usage
python3 scripts/build_workflow_templates.py --type casework --name "Citizen Housing Request" --output out/
# Custom research workflow
python3 scripts/build_workflow_templates.py --type research --name "Competitor Intelligence" --steps "scope,source discovery,evidence capture,synthesis,review,publish" --output out/
Output Files
- <slug>.workflow.yaml — Machine-readable workflow contract with all steps, gates, and evidence requirements
- <slug>.playbook.md — Operator playbook with numbered steps and handoff guidance
- <slug>.test-plan.md — Validation scenarios: happy path, missing input, approval gate, external failure escalation
Requirements
- OpenClaw v2026.3.23 or later
- Python 3.8+ (stdlib only — no pip installs required)
- No API keys