Verify What Matters: The 'test' Skill for Agent-Driven Proof

Where this fits

You just finished a branch that changes login behavior. The build is green, but you need to prove that a user can still sign in, see the dashboard, and hit the new error path. The test skill takes that diff, branch, PR, or URL and maps it to acceptance criteria, then runs focused checks plus a real browser for browser-facing work. It is for developers who want evidence, not just a passing pipeline.

Why agents benefit

• It preserves acceptance criteria wording and IDs (AC-1, INV-2) instead of rewriting tests to match whatever the code currently does.
• It maps every criterion, invariant, and failure path to a specific proof, so missing coverage becomes explicit.
• It knows when to add or update focused automated tests and when to explain why a code change cannot be proven by tests alone.
• It asserts observable user or caller behavior, which catches regressions that source-code reading misses.
• It can drive a real browser for end-to-end user flows, making agent verification useful on UI-heavy work.

Practical scenarios

Testing a PR against its acceptance criteria

An agent reads the PR description, changed files, and existing test suite. It adds a focused test for the new behavior and runs it in a browser to prove the flow.

Verifying a refactor preserved invariants

A developer refactors payment retry logic. The skill identifies behavior a refactor must preserve, then runs targeted tests around failure paths and retries rather than the full suite.

Checking a deployed URL or user flow

A designer reports a broken onboarding step. The agent uses the skill's real-browser support to walk the exact flow and report which criterion fails.

Add it to your agent workflow

Install the skill from the blueprint repo:

npx skills add owainlewis/blueprint --skill test

Then ask your agent to verify work by pointing it at a task and acceptance criteria:

Test branch `feature/password-reset` against AC-1 and AC-2.

The agent will inspect the change, add proof where missing, and report evidence by criterion.

Compare before adopting

Before adopting test, compare these signals:

• Quality signals: Does it preserve acceptance criteria IDs? Does it distinguish “green build” from “behavior proven”?
• Maintenance freshness: Is the repository active and does the skill keep up with current agent tooling?
• Alternatives: General testing prompts or standalone browser tools may lack the mapping from criteria to proof.
• Workflow fit: Does your agent already have browser access? Will the skill slot into your existing branch/PR workflow?

Why it is worth tracking

With 375 stars and a focused README, test is a useful addition to the open agent-skill ecosystem. Its value is not in running every check; it is in turning acceptance criteria into explicit proof, which is the difference between a passing build and a verified behavior. Evaluate it when your agents need to close the loop on code changes with browser-verified evidence. You can track it and related skills via OpenAgentSkill to see how it evolves.