adventurewave-labs

We build open-source developer tooling for the Claude and agentic AI ecosystem.

Founded by Marcus Patman — creator of Turbo Flow (versions 1–4 were a full multi-agent dev environment; v5 is the rules layer for AI-written code) and Turbo Rig, the lab's current agentic coding rig.

What is Adventure Wave Labs?

Adventure Wave Labs (AWL) is the open-source lab behind Turbo-Flow and its supporting toolchain — CLIs, agentic loop runners, and code-intelligence engines that let a single engineer operate a full agentic development workflow on top of Claude Code. Everything here is Rust-first, MCP-native, and built to run in real terminals against real codebases — not demos.

The Turbo Rig Stack - A Complete Agentic System

The successor to Turbo Flow — not more agents, but the harness that governs them: Triangle + Spine + Loop. Swappable lanes (builder / reviewer / reserve) under one shared constitution, a review gate cross-checked across model families with fail-closed verdicts, worktree isolation for parallel writers, git-versioned cross-session memory — and the merge button always in human hands.

Tool	Stars	Lang	Purpose
turbo-rig	—	Shell / Python	The full rig: a ten-law constitution, `review.sh` — the review gate with token accounting, `wt.sh` worktrees, `secret.sh` keychain, git-versioned `memory/`, specs and runbooks. One command installs and wires it all. Private repo — beta.
Turbo Brain v2	—	Python / Shell	The rig's memory plane — an owned context vault: plain markdown in git, served read-only over MCP so a session starts with the context it needs instead of asking for it. Capture is a pure text PUT; schema lint, secret scan, client deny-list and agent-poison tripwires gate what enters, and budget-capped distill waves land as PRs under a protected main. Private repo — v1.2.0.
turbo-rig.com	—	—	The product page: the full methodology — why builder ≠ reviewer, the three execution planes (laptop / VPS / Codespaces), and the eight-station loop that ends in a human merge.
Deep-dive	—	—	Technical walkthrough with 11 SVG diagrams — triangle, spine, planes, and loop, section by section.
Deep-dive 3D	—	—	The same architecture as an explorable 3D scene — the loop, the triangle, the gate, the data gravity well, the planes, and the automation ring.
7-Day Stats	—	—	One engine, Sept 14–21, 2026: 141 PRs merged across 5 repos, 868 gate verdicts (79% of them REVISE), 104 worktree lanes, 4.21B tokens, $416 total review spend.
Beta	—	—	Request access to the private repo during the beta.

The Turbo-Flow Stack - The Rules Layer for AI-Written Code (v5)

Tool	Stars	Lang	Purpose
turbo-flow		Shell / Python	Versions 1–4 (2025–2026): full agentic dev environment — 215+ MCP tools (via Ruflo v3.5 by ruvnet), cross-session memory (Beads), codebase knowledge graph (GitNexus), per-agent git-worktree isolation, one-command bootstrap on DevPod, Codespaces, or Rackspace Spot. Since v5: rig-lite — a portable bash-only governance kit (fail-closed cross-model review gate, constitution, git-versioned memory) for any repo and any coding agent. Site: turboflow.space · Academy: Turbo Flow University
turbo-flow-wizard		Shell	Guided setup wizard for turbo-flow — interactive generator for project-specific CLAUDE.md configs. 12 app types, 7 methodologies, 19 feature sets.
loopgen		Rust	Agentic loops for Claude Code — wizard, TOML configs, bash export, LOOP_STATUS protocol. Published on crates.io.
tf-verify.sh	—	Shell	50+ quality gates across 12 verification phases — dependency integrity, deployment state, artifact validation, and environment checks. Bundled in turbo-flow `devpods/`.
turbo-brain-v2	—	Python / Shell	The stack's knowledge vault — the same Turbo Brain v2 listed in the Turbo Rig Stack table above; one repo, serving Claude Code, Cowork and Turbo Flow over read-only MCP (`brain_search`, `brain_context`, `brain_verify`). Private repo — v1.2.0.

In motion


turbo-flow — real install, real tmux workspace, real claude launch	turbo-flow-wizard — interactive CLAUDE.md generator


loopgen — agentic loops for Claude Code: wizard, TOML, bash export

What We Build

DevOps / Infrastructure

Autonomous agents, reliability tooling, and infrastructure posture scanning. These projects sit between your CI/CD pipeline and your fleet — they watch, diagnose, heal routine failures, and gate deployments on cluster posture without human intervention.

ansible-heal-agent heals 3 seeded failures (stale hostname, removed apt_key, undefined nginx_port) and lands 3 conventional commits in 7 seconds.

Repo	Lang	What it does
ansible-heal-agent	Python	Autonomous agent that scans Ansible logs, diagnoses routine failures (stale hostname, removed module, undefined variable), patches the playbook / inventory / vars, commits via conventional commits, and re-runs the pipeline. LLM-first (GLM-4-Plus via `z-ai-web-dev-sdk`) with a deterministic rule-based fallback, YAML validation before write, and a full Markdown transcript for human audit. `make demo` heals a broken baseline end-to-end in about 7 seconds.
noip-scanner	TypeScript	NOIP — read-only Kubernetes posture scanner. 15 deterministic checks (pod security, NetworkPolicy, RBAC, CIS L1 workload subset) with NSA/CISA and NIST reference mappings; live cluster, offline manifest, or multi-context fleet scans; JSON / MD / SARIF / HTML reports in EN/ES; `--fail-on` pipeline gating; audit evidence bundles with DSSE signing; ValidatingAdmissionPolicy generation. Golden-tested against a real kind cluster in CI.

In motion

noip-scanner scanning a seeded kind cluster — live scan (score 35/100, 17 findings), --fail-on exiting 2 as a pipeline gate, evidence bundle + verify-bundle, ValidatingAdmissionPolicy generation, and a Spanish report.

Developer Tooling

CLIs and engines that plug into the agentic workflow — an agentic loop runner, a CI secret scanner, a code-intelligence engine, and a pre-deploy test harness. Each one shown doing real work against a real target in the grid below.

Repo	Lang	What it does
loopgen-rs	Rust	Agentic loop runner for Claude Code — compiles a goal into a structured harness and drives `claude -p` around a PLAN → ACT → VERIFY → REPORT cycle until a parsed `LOOP_STATUS` contract trips: `DONE` (optionally gated on a real verify command), `BLOCKED`, or a hard `--max` iteration cap
secret-scan	Rust	Regex-based secret scanner for CI pipelines
codescope	Rust	Single-binary code intelligence engine for AI coding agents — tree-sitter, MCP, CLI
preflight-integration-tester	Python	Pre-deploy integration test harness

In motion

loopgen-rs — agentic loop runner	codescope — code intelligence for agents
secret-scan — CI secret scanner	preflight-integration-tester — AI readiness diagnostic

Lab / Demos

Experiments, proofs of concept, and agentic demos.

Repo	What it explores
agentic-powered-golden-path-demo	NL → GitOps deployment via golden-path workflows (ArgoCD + OpenRouter)
AI-Kubernetes-API-Generator-Demo	NL → Kubernetes API generation
agentic-devops-extravaganza · live demo	Working demo of K8sGPT and Robusta running against a real Kubernetes API + GLM-4.5 LLM
agentic-platform-engineering-extravaganza · live demo	Golden path + Score → Crossplane + real OPA policy gates + an authz-gated MCP server — 42 policy violations without a platform, 0 with one
aops-sre-pipeline · live demo	Alert-driven autonomous SRE pipeline — Prometheus → n8n → Popeye → Dify-lite → Ollama → Slack
gitops-progressive-delivery-demo · live demo	Argo CD + Argo Rollouts + Prometheus + K8sGPT progressive delivery — a canary trips an SLO violation, an AI SRE diagnoses it, and the rollout auto-rolls back in ~20s
moor · live demo	Desired-state control plane for docker-compose — OBSERVE → DIFF → PLAN → ACT drift detection with real auto-remediation against the live Docker Engine API
cloudtrim · live demo	22-rule AWS cost-waste audit & remediation engine — read-only scan, verified fixes, and CUR-reconciled savings reporting demoed against a pinned AWS API emulator

FAQ

What is Adventure Wave Labs? An open-source lab building developer tooling for the Claude and agentic AI ecosystem — CLIs, agentic loop runners, and code-intelligence tools used alongside Claude Code.

What is Turbo-Flow? Turbo-Flow is Marcus Patman's open-source project (MIT). Versions 1–4 (2025–2026) were an agentic development environment — 215+ MCP tools, cross-session memory, per-agent git-worktree isolation. Since v5 it is the rules layer for AI-written code: rig-lite, a portable bash-only governance kit that works in any repo with any coding agent. See marcuspat/turbo-flow.

Is this tooling used in production? The toolchain is dogfooded daily inside Turbo-Flow, and most repos have their own CI (see each repo's Actions badge). preflight-integration-tester is explicitly marked pre-alpha in its own README — it's a real, working tool, just not yet positioned as a production dependency for outside teams.