What is the purpose of 'prove'?

To provide a concise, verifiable demonstration that the work satisfies the original request, with reproducible steps to verify.

How do I verify the proof?

Follow the included verification steps: run cli commands, open URLs, inspect diffs, and compare outcomes to acceptance criteria.

What if the proof reveals gaps?

Iterate between implementation and proof, update the plan, re-run verification, and use an oracle to ensure alignment.

prove

Scanned

npx machina-cli add skill bryanp/agent-skills/prove --openclaw

Files (1)

SKILL.md

758 B

Generate proof that the work you implemented satisfies the original request.

Rule: Content

The proof should be as concise as possible. It should thoroughly demonstate to a human how the implementation works and how they can verify it works. Provide steps they can take to run and verify the implementation themselves (cli commands, clicking through in a web browser, etc).

Approach this like you would a presentation to your boss.

Rule: Verify Proof

Iterate between implementation and proof until you're sure the implementation is correct. Use the oracle to compare the implementation to the plan and proof so that the conclusion is unbiased. Err on the side of correctness!

Source

git clone https://github.com/bryanp/agent-skills/blob/main/code/prove/SKILL.mdView on GitHub

Overview

Prove creates a concise, verifiable justification that your work satisfies the original request. It guides you to show exactly what was built, how to test it, and how to compare results against the plan. The output is designed to be self-contained and auditable by a reviewer.

How This Skill Works

Generate a short, concrete proof that links the implementation to the stated goals. Include concrete verification steps (CLI commands, browser checks, diffs) so a reviewer can reproduce and confirm results. The process often involves iterating with an oracle to align implementation and proof.

When to Use It

After implementing a feature to demonstrate it meets the requirements.
During code reviews to show alignment with the plan.
Before client handoff to provide verifiable evidence of correctness.
During refactoring to ensure no regressions were introduced.
When validating complex outputs against acceptance criteria.

Quick Start

Step 1: Define success criteria and plan the proof.
Step 2: Implement the solution and collect reproducible evidence (tests, screenshots, CLI outputs).
Step 3: Compile and share the proof, then verify against the plan (iterate if needed).

Best Practices

Define clear success criteria before implementation.
Keep the proof concise yet complete enough for verification.
Provide reproducible steps to verify (CLI, URLs, diffs).
Use an oracle or peer review to confirm alignment.
Update the proof whenever the implementation changes.

Example Use Cases

Prove a REST API endpoint returns the expected JSON schema.
Verify a data transformation produces correct input-output mappings.
Demonstrate a CLI tool's commands and outputs match the spec.
Show UI behavior via browser steps and expected states.
Confirm an AI agent's response adheres to constraints and style guidelines.

Frequently Asked Questions

Add this skill to your agents