principle-prove-it-works

Verifies completed work by directly inspecting real artifacts instead of proxies or self-reports.

Updated Sep 2, 2026
One-click install
npx skills add https://github.com/jnyross/pstack-muse --skill principle-prove-it-works-jnyross
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: principle-prove-it-works
Source: https://github.com/jnyross/pstack-muse/tree/main/skills/principle-prove-it-works
Command: npx skills add https://github.com/jnyross/pstack-muse --skill principle-prove-it-works-jnyross

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Work is often declared done based on indirect signals like successful compilation, file timestamps, or an agent's self-report, which can hide real failures. This Skill enforces direct verification against the actual artifact before any task is considered complete. ## Core Features & Use Cases - Direct Verification Discipline: Run the actual feature, read the actual value, and inspect the real diff rather than trusting derived state or cached output. - Delegation Auditing: When reviewing delegated or agent-produced work, inspect git diffs, file contents, and runtime behavior instead of trusting summaries. - Scripted Proof: Write deterministic verification scripts that re-run the same comparison, producing artifacts a reviewer can re-execute. - Use Case: After an agent claims a bug fix works, apply this Skill to run the feature end-to-end, confirm data flows from input to output, and keep the verification script output visible for review. ## Quick Start Apply the prove-it-works principle to verify my last change by running the actual feature and inspecting the real output before marking it done.

Frequently Asked Questions about principle-prove-it-works

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify that a code change actually works?

Build the code, then run it and exercise the actual feature path end-to-end. Check that data flows from input to output, and for integrations test the full communication path rather than stopping at a successful compile.

How to verify work delegated to an AI agent?

Trust artifacts, not self-reports. Inspect the actual git diff, file contents, and runtime behavior instead of the agent's summary, since agents report what they intended rather than always what happened.

Why is 'it compiles' not enough to declare a task done?

Compilation is necessary but not sufficient proof of correctness. Indirect signals like build success, file mtimes, or cached output can hide broken behavior, so the real feature must be exercised directly.

When should I write a verification script instead of checking manually?

Write a deterministic script whenever the same comparison can be re-run, such as comparing old and new compiled output. Keep the script's output visible as an artifact a reviewer can re-execute instead of trusting your word.

What should I suspect first when verification fails?

Suspect the observation method before suspecting the system. Check process liveness directly rather than through derived state, and read actual values rather than cached or derived representations.