FalseGreen installation

Connect Codex or Claude Code to independent acceptance.

Account. Self-serve Job. Durable result.

Define what done means, let the coding agent implement, and use FalseGreen to verify the exact result against the frozen boundary.

Learn how to define a good Job →
On this page

First-time setup

Install once. Connect Codex or Claude Code.

The normal path starts with an Account. If you bought a Job first, create or sign in to that account now so the purchase can link to you.

  1. 01

    Create your Account.

    Start the FalseGreen Account flow. After checkout, open your purchase or account from the confirmation and copy your FalseGreen key. The key is not included directly in the confirmation email.

  2. 02

    Install and log in.

    cargo install falsegreen
    falsegreen login --token YOUR_FALSEGREEN_KEY

    The key is exchanged for a local session. Do not put the key in your editor’s MCP configuration.

  3. 03

    Install the integration.

    Choose the exact command for the coding agent you use:

  4. 04

    Fully restart the coding agent.

    Quit the running application or process after installation, then reopen it. The FalseGreen tools may not appear until the coding agent has fully restarted.

    Ask the coding agent to list its available MCP tools before starting a Job.

The build and acceptance roles

The coding agent builds. FalseGreen decides whether it is done.

FalseGreen’s tools guide planning, freeze the acceptance boundary, run independent verification, and retrieve the report. Any MCP service in the software being built is part of the implementation under test, not the FalseGreen verification system.

Planning

Define the target before implementation.

Start by telling the connected coding agent the consequential engineering outcome. FalseGreen guides the plan through the objective, constraints, assumptions, required behavior, evidence, and failure conditions.

Frozen boundary

Freeze what done means.

FalseGreen validates and freezes an immutable acceptance boundary before the result is judged. Material changes require a new boundary instead of silently moving the target after the work.

Running a Job

One sequence from intent to evidence.

  1. Buy a Job before or after creating your Account.
  2. Tell the connected coding agent: “Use FalseGreen for this task.”
  3. Plan, validate, and freeze the acceptance boundary.
  4. Let the coding agent implement against the assignment.
  5. Run completion verification and inspect the FalseGreen result.

Buying the Job creates one Job entitlement. It is consumed when the first durable Job run begins, not when checkout completes.

View the tool sequence
falsegreen_create_task(workspace, goal, title)
falsegreen_get_plan_schema()
falsegreen_save_plan_draft(task_id, plan)
falsegreen_validate_plan(task_id)
falsegreen_freeze_plan(task_id)
falsegreen_get_plan(task_id, frozen=true)
falsegreen_save_contract_draft(task_id, contract)
falsegreen_validate_contract(task_id)
falsegreen_freeze_contract(task_id)
falsegreen_get_assignment(task_id)
# coding agent implements against the assignment
falsegreen_check_completion(task_id)
falsegreen_get_verification_artifacts(task_id, run_id)

Results

Read the server-issued result, not the agent’s confidence.

A Job can conclude Accepted, Failed, or Insufficient Evidence. The durable report can include the verdict, observed source identity, criterion-level evidence, limitations, and repair feedback.

Accepted means the frozen criteria were established by the evidence observed for that run. It is not a universal correctness guarantee or a certification of the model or codebase.