Goal to Proof
Open-source Agent Skill

Finish the work.
Prove the outcome.

Goal to Proof is a lightweight closure contract for AI agents: turn authorized, non-trivial work into an observed outcome—not a plausible completion claim.

Install Goal to Proof Read the method 
npx skills add aiopshwang/goal-to-proof
Agent Skills
Scope-matched proofEvidence is as broad as the claim.
User-owned authorityNo silent scope or permission expansion.
Low ceremonyProcess appears only when closure needs it.
Definition

What is Goal to Proof?

Goal to Proof is an AI agent completion-verification skill. It prevents an agent from substituting a plan, placeholder, partial artifact, isolated check, or unobserved external action for the requested result.

01 · ArtifactSomething exists

A file, change, or record was created.

02 · Focused checkOne layer passes

Syntax, unit test, citation, or schema is valid.

03 · IntegrationParts work together

The connected workflow runs.

04 · TargetThe real boundary works

The user, device, render, account, or remote sees it.

05 · OutcomeThe claim is observed

Stop at the first layer that directly proves the request.

Closure contract

Four fields. One honest finish.

The fields stay internal when obvious and become visible only when ambiguity or authority could change the work.

R

Result

What must be different when the task is done?

T

Target

Who or what must be able to use or observe it?

P

Proof

What direct observation separates success from plausible failure?

B

Boundaries

What is authorized, excluded, or requires the user?

Responsibility

Autonomy without taking over.

The user and agent have different jobs. Making the split explicit prevents both approval theater and unauthorized action.

The user owns

  • Goals and values
  • Authority and risk tolerance
  • Material product choices
  • External and irreversible decisions
  • Final judgment

The agent owns inside the boundary

  • Method and sequencing
  • Reversible implementation choices
  • Diagnostics and evidence gathering
  • Authorized execution
  • Verification and precise reporting
Activation

Use it where “done” can be misleading.

Good fit

  • Dependent, multi-step deliverables
  • Features and fixes with a real user path
  • Deployment or publication with remote state
  • Documents and visuals that need final rendering
  • Research with material, sourceable claims
  • Evidence-backed diagnoses, decisions, or executable plans as final deliverables

Not an implicit fit

  • Simple questions and direct answers
  • Translation or formatting
  • Open-ended ideation without a defined deliverable
  • Routine self-contained edits with an obvious direct check
  • Read-only answers that do not exercise a target workflow
  • Generic discovery or autonomous goal ownership
Install

One canonical skill, three distribution paths.

The repository root is the plugin root. Platform packages point to the same canonical skills/goal-to-proof/SKILL.md.

Agent Skills

Portable installer for supported skill hosts.

npx skills add aiopshwang/goal-to-proof

Codex marketplace

Add the repository catalog, then install the plugin.

codex plugin marketplace add aiopshwang/goal-to-proof
codex plugin add goal-to-proof@goal-to-proof

Claude Code marketplace

Install as a namespaced managed plugin.

claude plugin marketplace add aiopshwang/goal-to-proof
claude plugin install goal-to-proof@goal-to-proof
Explicit invocation: use $goal-to-proof in Codex. In a Claude Code managed plugin install, use /goal-to-proof:goal-to-proof; a standalone skill may be exposed as /goal-to-proof.

Package presence is not a blanket compatibility claim. See validation evidence and claim levels for what each check proves.

Origin

Distilled from real work. Published without private work.

The initial contract came from aggregate analysis of prior real working sessions. Repeated behaviors—not private content or corpus metadata—became the public principles.

Patterns retained

Intent alignment, scope control, authorized autonomy, end-to-end proof, durable checkpoints, and evidence-first reporting.

Material excluded

Raw conversations, personal names, secrets, one-off preferences, private content, and session transcripts.

Claim boundary

The corpus counts explain the source of the design. They are not a benchmark or proof of performance.

Read the methodology →

Quick answers

Frequently asked

Does it give the agent more permission?

No. Goal to Proof makes the current authorization boundary more explicit. Publication, spending, disclosure, irreversible action, and material scope expansion still require user authority.

Is this just “run the tests”?

No. Tests can be excellent proof for a software claim, but the requested boundary might be a rendered document, a public release, an account state, or a research conclusion. Proof must match the claim.

Does it guarantee completion?

No. Tools, permissions, environments, and the task itself can block completion. The skill requires the agent to report the verified layer and name what remains unverified.

Is Goal to Proof an SEO or GEO product?

No. It is an AI-agent completion skill. This site uses conventional search fundamentals and clear, source-linked answers, but the project makes no ranking claim. See the SEO/GEO answer.

Read all questions →