dacli / product pitch
The control plane for coding-agent swarms

Coding agents need
a delivery system.

dacli turns agent work into a durable record of what was planned, built, verified, and landed.

PlanExecuteVerifyReviewLand
The buyer & the problem

The engineering lead becomes
the coordination layer.

One agent writes code. A team of agents needs ownership, shared context, review, and a way to know what actually shipped.

Before the run

Who owns each file? What must this task prove?

During the run

Which worker is blocked? What budget remains?

After the run

Did the reviewed change reach the intended branch?

The first user: maintainers and engineering leads operating coding agents.
The product

Let agents judge.
Make the process checkable.

Human

Sets direction, credentials, exceptions, and release authority.

dacli

Enforces scope, budgets, state transitions, recovery, and delivery evidence.

Coding agents

Plan, implement, test, and review through the selected coding harness.

CLI and MCP access the same durable workspace.Architecture
Working product

The first proving ground
is dacli itself.

dacli is developed using dacli.

Task attribution, isolated worktrees, and a reproducible empty-repository-to-shipped-code fixture make the delivery process inspectable.

The deterministic fixture proves coordination. It is not a benchmark of model quality or customer traction.

Public engineering evidence, with the verification boundaries attached.
See the product

A shared view of progress.
Evidence behind every state.

Representative dacli dashboard showing outcome analytics, evidence coverage, and delivery state
Representative test fixture. These values are illustrative, not live operations or traction.
The workflow

One task. A chain of evidence.

  1. Select

    Ready work, dependencies, and critical-path priority.

  2. Isolate

    A scoped brief, worktree, path claim, and bounded run.

  3. Verify

    Repository checks and observable acceptance criteria.

  4. Review

    An independent verdict bound to the exact tree.

  5. Land

    GitHub checks and observed merge state before completion.

Where dacli fits

Keep the coding harness.
Add delivery control.

Different roles in the same workflow
Layer Its job
Codex, Claude Code, Gemini CLI, Copilot CLI Provide the coding-agent runtime.
GitHub issues, pull requests, and CI Provide collaboration, checks, and merge evidence.
dacli Coordinate the task lifecycle across agents and preserve its evidence.

A run stays on its selected harness. Cross-harness routing requires an explicit allowlist and hybrid configuration.

The adoption thesis

Start with one repository.
Earn the next run.

The initial wedge is an engineering team already using coding agents, with a backlog and a delivery process worth preserving.

Try a bounded task

Declare the acceptance criteria and inspect the recorded result.

Repeat with a bounded loop

Expand only after scope, review, cost, and recovery are understood.

The product boundary

An open core today.
A larger operating layer ahead.

Available now

The open-source repository workspace, CLI, MCP server, dashboard, and bounded orchestration workflows.

BSD-3-Clause license

Expansion direction

A hosted control plane for shared operations across repositories and tenants. The repository documents its implementation and deployment boundaries.

This is a product direction, not a claim of a generally available hosted service or paid plan.

Put the process to work

Bring one real task.
Inspect what ships.

Start with the dashboard. Read the operating model. Run a bounded task on a repository you know.