venutian-antfarm

Getting Started with Venutian Antfarm

Part of Venutian Antfarm by RD Digital Consulting Services, LLC.

A step-by-step guide to setting up your agent fleet harness. For background on the framework’s philosophy, see Governing the Ant Farm.

Prerequisites

• Claude Code CLI installed and configured
• Git for version control
• gomplate v4+ for compliance floor compilation
• jq for metrics processing (optional but recommended)

Setup Overview

flowchart LR
    S1["1. Clone"] --> S2["2. Compliance\nFloor"]
    S2 --> S3["3. First\nSpecialist"]
    S3 --> S4["4. Fleet\nConfig"]
    S4 --> S5["5. First\nIteration"]
    S5 --> S6["6. DORA\nDashboard"]
    S6 --> S7["7. Fleet\nLearns"]

    style S1 fill:#bdbdbd,stroke:#424242,color:#1a1a1a
    style S2 fill:#ef9a9a,stroke:#b71c1c,color:#1a1a1a
    style S3 fill:#a5d6a7,stroke:#2e7d32,color:#1a1a1a
    style S4 fill:#ffcc80,stroke:#e65100,color:#1a1a1a
    style S5 fill:#a5d6a7,stroke:#2e7d32,color:#1a1a1a
    style S6 fill:#90caf9,stroke:#1565c0,color:#1a1a1a
    style S7 fill:#a5d6a7,stroke:#2e7d32,color:#1a1a1a

Step 1: Clone the Template

git clone https://github.com/rdunie/venutian-antfarm.git my-project
cd my-project

Step 2: Define Your Compliance Floor

The compliance floor is the set of non-negotiable rules that every agent must follow, regardless of pace or autonomy level. It encompasses security, data governance, audit requirements, regulatory controls, and domain-specific rules. Floors live in the floors/ directory at the project root; the compliance floor is guarded by the CRO.

mkdir -p floors
cp templates/floors/compliance.md floors/compliance.md

Edit floors/compliance.md with your domain’s rules. Keep it to 3-5 rules that are absolute, enforceable, and clear. For example:

• An e-commerce app might include: “No credit card numbers stored outside the payment processor”
• A healthcare app might include: “No PHI transmitted without encryption”
• A SaaS platform might include: “Tenant data isolation enforced on every query”

Step 3: Add Your First Specialist Agent

Copy a template and customize it for your tech stack:

cp templates/agents/backend-specialist.md .claude/agents/backend-specialist.md

Edit the agent file to match your domain. The key sections to customize:

• Domain: What this agent owns (e.g., “Django REST API, PostgreSQL models, Celery tasks”)
• Responsibilities: Specific to your project
• Autonomy Model: Start conservative, expand as trust builds

Agent Inheritance

Your specialist agents can extend the harness core agents. Add an extends field to the frontmatter:

---
extends: harness/scrum-master
---
# Your overrides here. For example, change retro cadence:
## Process Override
Retro cadence: every 2 items instead of every 1.

When extends is present:

• Fields in your app agent override the same fields in the harness agent
• Fields not mentioned in your app agent are preserved from the harness agent
• This lets you keep the full collaboration protocol while customizing behavior

Step 4: Configure Your Fleet

cp templates/fleet-config.json fleet-config.json

Edit fleet-config.json:

• project.name: Your project name
• metrics.backend: Start with "jsonl" (works out of the box)
• agents.specialists: List your specialist agent names
• retro.cadence: How many items between retrospectives

Step 5: Run Your First Iteration

Open Claude Code in your project directory. The 13 core agents are ready (7 governance + 6 operational):

Available slash commands:

Command	Purpose
/po	Product owner status and backlog
/retro	Sprint retrospective
/onboard	Fleet onboarding (activates CRO and CISO during setup)
/compliance	Compliance program management
/governance	Executive governance
/pace	Pace control
/audit	Compliance audit
/memory	Knowledge management
/deploy	Deployment orchestration
/findings	Findings register
/handoff	Structured handoffs

Rewards system: Use ops/rewards-log.sh to issue behavioral feedback (kudos, reprimands, tensions) and query agent profiles. See ops/rewards-log.sh --help for usage.

1. Start with /po triage to prioritize the backlog (Phase 0 Prioritize – triage before grooming)
2. Add a work item to docs/plans/ – even a simple one like “Set up project structure”
3. Run /po groom to have the PO add acceptance criteria and WSJF score
4. Build it – the specialist agent handles: code, test, typecheck, build, deploy, validate
5. Run /po review to verify against acceptance criteria

Step 6: See Your DORA Dashboard

After completing a few items:

ops/dora.sh           # full DORA + flow quality dashboard
ops/dora.sh --sm      # pace recommendation
ops/dora.sh --flow    # handoff quality by boundary pair

The dashboard shows deployment frequency, lead time, change failure rate, first-pass yield by handoff boundary, rework cycles, and more.

Step 7: The “Aha Moment”

As the fleet works through items, you will notice:

• Findings decrease. The same type of mistake stops recurring because the SM triggers knowledge distribution, and knowledge-ops executes under CKO direction.
• Handoffs get cleaner. Agents learn what their counterparts need. FPY improves.
• Pace naturally accelerates. When CFR drops below 10% and FPY exceeds 80%, the SM recommends advancing from Crawl to Walk.
• Specialists make good decisions autonomously. The PO and SA invest in context enrichment, and it pays off – specialists stop asking for every judgment call.

This is the fleet learning autonomy. Not because you told it to, but because the structured learning loop (findings, curation, refinement, distribution) compounds over time. Each iteration makes the next one smoother.

Next Steps

• Read the Agent Fleet Pattern for the full specification
• Review the Collaboration Protocol for all 11 core principles
• Check the Examples for progressive working examples (start with 01-getting-started)
• Add review agents (security-reviewer, etc.) as your compliance needs crystallize. Note: governance tier agents (CRO, CISO at minimum) are active from the first session via /onboard and don’t need to be added separately.

This site is open source. Improve this page.