Local audit for agent-written tests

Glubean Extension
for agent-written tests.

Open the test your agent wrote, run it inline, inspect the trace, debug real TypeScript, and decide whether the output deserves to become repo-owned verification.

Best with agent-written testsFree foreverNo account requiredRuns locally firstOutput stays yours
Read the agent guide
TypeScript test file with inline play button and live trace viewer
FocusInline trace
TypeScript test file with inline play button and live trace viewer
VS Code debugger paused at a breakpoint inside an API test
Side-by-side diff of two API trace runs
Test results panel showing 4 tests passed in 3.1 seconds
Environment switcher showing default, production and staging environments

Click play to run. The trace opens beside the test immediately.

Local audit proof

Turn agent output into something you can inspect.

The extension turns agent-written tests into normal engineering work: code on the left, evidence beside it, debugger when needed. `explore/` is where exploratory API work can replace Postman without trapping drafts in a second tool.

Run the draft inline

Open the test your agent wrote and run it from the gutter without leaving the editor.

Debug real TypeScript

Set breakpoints, inspect variables, and step through the code instead of guessing from terminal output.

Inspect traces and diffs

Compare runs, inspect evidence, and decide whether the generated test is trustworthy enough to keep.

Promote the same file

A test can start in `explore/` as a Postman replacement, then become the same file you commit, review, and promote to CI.

End to end

From `explore/` draft to owned verification.

The extension is where you validate what the agent produced and decide whether the draft earns a place in the repo.

Prompt and context

Agent prompt

“Verify checkout with promo code, tax calculation, and webhook delivery.”

Context fed into generation

context/openapi-open.json.claude/rules/glubean.mdtests/configure.tsmcp://glubean/run_local_file

Agent draft

explore/github.test.ts
import { test } from "@glubean/sdk";

export const listUserRepos = test(
  { id: "github-list-repos", tags: ["explore"] },
  async (ctx) => {
    const username = ctx.vars.require("GITHUB_USERNAME");
    const token = ctx.vars.get("GITHUB_TOKEN");

    const headers: Record<string, string> = {
      Accept: "application/vnd.github+json",
    };
    if (token) {
      headers["Authorization"] = `Bearer ${token}`;
    }

    const data = await ctx.http
      .get(`https://api.github.com/users/${username}/repos`)
      .json();

    ctx.expect(data.length).toBeGreaterThan(0);
    ctx.log("Repos", data.slice(0, 3));
  },
);

What the agent did

Used your OpenAPI routes

Matched auth, cart, promo, and checkout paths from the project spec.

Generated evidence-backed SDK code

Named steps and typed assertions, not free-form fetch calls.

Followed project conventions

Respected existing config helpers and file placement.

Ready to run immediately

Click play to see traces and verify the draft before committing it.

Starts where Postman used to live

Exploratory API work can begin in explore/, then graduate into durable tests without changing tools.

Prompt

Context goes in

Code

TypeScript comes out

Trace

Evidence closes the loop

Why the draft stays grounded

Agent output still needs guardrails.

The extension feels good because the draft is not generic. Specs, skills, SDK surfaces, and MCP tools keep the test tied to your project instead of drifting into demo code.

Targets the evidence SDK

Agent-written code uses named steps, typed assertions, shared config, and traceable requests instead of free-form fetch scripts.

Learn more

OpenAPI context stays focused

Annotated specs and reduced context keep the agent focused on the routes, constraints, and response shapes that matter.

Learn more

Skills lock what matters

A skill file teaches the agent your exact SDK API, auth patterns, assertion style, and conventions so drafts follow your team's patterns from the start.

Learn more

MCP closes the loop

MCP tools let the agent run tests, read structured failures, fix code, and rerun without dropping back to pasted logs.

Learn more

Inside the extension

Run, inspect, debug, diff. Stay in one window.

Once the agent writes the draft, the extension keeps the feedback loop tight: run it, check the trace, fix what needs fixing, and move on without leaving the editor.

Inline play buttons

Run one test, one file, or the whole workspace directly from the editor gutter.

Breakpoint debugging

Step through real code and inspect variables instead of guessing from logs.

Trace diff

Compare two runs with native diff to see exactly what changed in behavior.

Schema and contract checks

Use built-in validation patterns so generated code has real assertions behind it.

Environment switching

Move across local, staging, and production context from the status bar without rebuilding the test.

Git-native files

Everything stays in TypeScript, so PR review, branching, and refactoring work like normal.

Best visual path

Install the extension. Review the test where it lives.

Install the extension when you want the fastest way to inspect, debug, and defend agent-written verification. The output is still real TypeScript that stays in your project.