An evidence harness for Codex and Claude Code · native macOS
Stop steering the agent. Start judging the result.
The agent writes the code. GoalCoder runs the process and shows exactly what happened — repository reconnaissance, the questions that matter, validation, review, and a handoff with the evidence attached.
Apple silicon Mac · your own Codex or Claude Code login · tax-inclusive USD 19 a month after the 14-day trial.
New Run. Define the outcome, repository, provider, and runtime in one native Mac workspace.Full resolution
Your workflow
The same ticket, twice.
Today
You steer the session
You paste the ticket, then spend the session steering:
“check the repo first” · “don't edit yet” · “what about the edge case” · “write tests” · “actually run them” · “show me the diff” · “what did you skip”
You cannot step away. At the end you still do not know what happened versus what it told you.
With GoalCoder
You judge the result
You state the outcome and answer only the questions it genuinely needs. What comes back:
The changed work and its diff
The validation that ran, and its output
Review findings, and the fixes made
Retained artifacts and a candid next action
Caveats it refused to guess about
You review it once, and decide.
What comes back
The handoff keeps caveats visible.
A completed run does not flatten partial proof into success. GoalCoder keeps the lifecycle, evidence, unresolved caveats, and next actions together for inspection.
Completed with caveats. The native result view separates accepted evidence from what still needs attention.Full resolution
The GoalCoder Method
One request, carried through the work that must be true.
GoalCoder packages the operating discipline that otherwise lives across a long chain of prompts, reminders, and manual checks. It raises the level of your request without dropping the rigour a repository with real conventions, tests, and reviewers demands.
01 · Request
Ground the goal in the repo
GoalCoder inspects the repository, its instructions, current state, and the boundary of the requested outcome before editing.
02 · Plan
Resolve what matters
A material Question Gate asks for decisions only when repository evidence cannot supply them. A reviewed plan defines scope and proof.
03 · Work
Implement against the plan
The provider works in a GoalCoder-owned edit workspace and follows the repository's real tools and constraints.
04 · Review
Test the result, not the story
Validation and review check the requested behavior, important failure paths, edge cases, and the final diff.
05 · Result
Return evidence and caveats
You receive the changed work, validation evidence, review outcome, retained artifacts, and a candid next action.
What it needs, and what it touches
Bring a Mac, a repository, and one provider.
GoalCoder stores and manages repository and run content on your Mac. Codex or Claude Code still connects to the provider you choose through your own account.
01
Apple silicon macOS
GoalCoder v1 is a native Mac application for Apple silicon. Windows, Linux, Intel Mac, and the Mac App Store are outside the v1 offer.
02
An existing repository
GoalCoder is designed for coherent, reviewable technical work in a repository you are authorized to use.
03
Codex or Claude Code access
One compatible, logged-in provider is enough. GoalCoder does not store provider credentials or require access to both providers.
Uses the installed AI agent on your Mac. Edits happen in a separate GoalCoder workspace. The provider still runs as your user account, so this mode provides no host isolation.
Stronger isolation
Sandboxed · Beta
Runs the AI agent inside Docker Sandboxes with a private filesystem, network, and Docker. Beta means a narrower tested range and more setup before the first run.
The 14-day trial is the opening period of the same full-feature subscription. After the trial, it renews automatically each month unless cancelled.
USD 19
per month · tax included
In addition to your existing Codex or Claude Code plan. GoalCoder sells the orchestrator, not model usage.
Full GoalCoder workflow and both provider adapters
Local trusted and Sandboxed · Beta runtime modes
One person on up to three Macs
Compatible application and workflow updates
No GoalCoder run or provider-usage meter
Provider subscriptions, limits, and charges are separate. Normal cancellation takes effect at the end of the current trial or paid billing period.
Questions that matter
Know the boundaries before you use it.
Does GoalCoder edit my selected checkout?
No. GoalCoder creates a separate edit workspace. Local trusted protects the selected checkout but does not isolate the provider from your Mac; Sandboxed · Beta provides the stronger host boundary.
Does GoalCoder send my code to its servers?
No. GoalCoder stores and manages repository and run content on your Mac. The app never sends that content to GoalCoder servers. Codex or Claude Code still has access to the repository through your own provider account.
Do I need both providers?
No. One compatible, logged-in Codex or Claude Code session is enough for runs using that provider. Your GoalCoder subscription includes both adapters, but it does not include provider access.
Does GoalCoder meter my work?
No. GoalCoder does not meter runs or provider usage. Provider plans, limits, token usage, and charges remain between you and your provider.
What happens when access ends?
New runs stop. Existing local runs, workspaces, artifacts, logs, and handoff evidence remain on your Macs for inspection.
How do cancellation and refunds work?
Manage or cancel your subscription through Link. The first time, Link may ask you to create an account. Product and account questions go to GoalCoder support; Link provides transaction support. Statutory consumer rights and accepted refunds continue to apply.
Judge it on your own repository
Run one real ticket through it. What comes back has already been questioned, validated, and reviewed.