Steer, build, and accept stay distinct.
An agent working inside a code branch cannot quietly weaken a test or rewrite its own requirements to make a build pass. Defining the goal, writing the code, and accepting the result are separated by design.
The Operating Agreement for AI-Assisted Teams
AgentFlow SDLC 2.0
Let AI write the code.
Keep engineering judgment, risk boundaries, and acceptance yours.
Fast code generation is only a win if your team still knows what shipped and why it is safe. AgentFlow 2.0 gives AI-assisted engineering teams a clear operating model: the agent that writes the code never approves the release, ceremony scales with risk, and recurring mistakes turn into shared team rules.
Open source (Apache-2.0) · Works with Claude Code, Codex, Gemini CLI, Grok, and OpenRig · Scales from one developer CLI to a full team squad
One slice. Proportional ceremony by risk.Click a scenario to see how governance adapts
01 / STEER & SCOPEFrameGoal & risk boundary
02 / LOCK CHECKSHand overPinned test criteria
03 / BUILDSingle-Pass BuildBuilder + helper sub-agents
04 / VERIFYCheck EvidenceTests & boundaries pass
05 / VERDICTAcceptHuman-owned release gate
When changes need rework, context stays intact instead of starting from scratch
Routine change ready for acceptance in one pass
Test criteria are locked up front. Low-risk work completes in a single builder pass without unnecessary bureaucracy.
Routine slice: automated test and file-boundary checks run first; no heavyweight review panel required.
Why care · What it brings to the table
AI agents are great builders.
They should never grade their own homework.
Built for VP Engineering, CTOs, tech leads, and developers who want the speed of agentic coding without losing accountability over their codebase.
Fast on routine work. Strict where it counts.
A simple UI or docs update finishes in a single pass. The moment a change touches authentication, billing, migrations, or security rules, AgentFlow automatically escalates to independent cross-model review.
Turn today's friction into tomorrow's guardrails.
When an agent gets stuck in a rework loop or hits a boundary violation, AgentFlow captures the pattern and proposes a clear, human-reviewed update to your repository's engineering rules.
Built with AgentFlow
See a real product.
Inspect how it reached production.
AwesomeAwesomeness was built across multiple sessions using AgentFlow's role boundaries, automated evidence checks, and human acceptance gates. Both the live application and its complete delivery history are public to inspect.
- 3delivery episodes
- 2clean handoff recoveries
- 100%verified release evidence
Adopt at your own pace
Start with one coding CLI.
Scale to a multi-agent squad.
You don't need a complex multi-agent cluster to start. AgentFlow uses the coding tools already installed on your machine and grows with your team.
See the 3-step setup →Solo developer, clear guardrails.
Run AgentFlow with a single tool (Claude Code, Codex, Gemini CLI, or Grok) to keep your requirements, test criteria, and release checks structured.
Independent "Four-Eyes" review.
Have two coding CLIs installed? Let one model family build the feature and a different model family review the diff—catching blind spots before you review.
Dedicated seats for larger teams.
Coordinate dedicated architecture, building, QA, and review seats working across isolated Git branches with shared governance.
Where it fits in the ecosystem
The governance layer of the Move the Needle stack.
AgentFlow SDLC 2.0 works completely standalone, and connects cleanly with the rest of our open-source systems when you want parallel sandboxes, live telemetry, or personal context.
Governance meets parallel sandboxes.
AgentFlow defines the scope, risk tier, and acceptance criteria; Meshloop executes the build tasks in parallel isolated Git worktrees.
Live dashboard + retrospective flight recorder.
Watch live agent behavior in Grafana or Datadog with agent-otel-bridge, and analyze session time, rework loops, and cost afterward with Dendro.
Your engineering principles in every repo.
Link repositories to Holoself so agents inherit your approved architectural standards and working preferences automatically.
Start small. Scale with confidence.
Try AgentFlow 2.0 on one repository today.
Built by Samuel Mota · Move the Needle
Why AgentFlow SDLC exists
Open source · Apache-2.0 · Release 2.0
Value
Let AI write the code while keeping engineering judgment, risk boundaries, and release acceptance explicitly human-owned.
Strategy
When code generation becomes cheap, decision quality and clear ownership become the bottleneck. AgentFlow keeps roles, risk boundaries, and release sign-off in your team's hands.
Technical frontier
Separates steering, building, and independent acceptance across risk-calibrated slices, turning recurring review friction into human-approved engineering rules.