Skip to content

One rule for trusting agent work across a team of twenty to fifty, whichever agent each engineer uses

Excellent gives every agent run the same shape: attempt, checks, evidence, result. Reviewers compare records instead of arguing from transcripts.

curl -fsSL https://excellent.so/install.sh | sh

What you recognise

  1. 01

    One engineer trusts a passing unit test, another wants browser evidence, and nobody can say what the agent actually observed.

  2. 02

    Agent runs happen in local terminals, pull requests and CI logs, and the evidence is scattered across all three.

  3. 03

    A regression gets through because the reviewer had to reconstruct the context after the fact.

What changes

One result vocabulary

Every run ends verified, returned or unchecked, with the same fields behind each word, so a review conversation starts from the same facts.

Evidence stays attached to the attempt

Repo state, command output, test results and reviewer notes belong to the run that produced them, not to whoever remembers them.

No self-approval

The worker's completion claim is kept separate from the evidence that decides whether the work moves forward.

See for yourself

  • Checks and results

    What a check is, and what each result state commits to.

  • Work and attempts

    The record shape a reviewer reads, including retries.

  • Teams

    What running Excellent across a team looks like today, and what it costs.

  • Security and data

    Where attempts, artifacts and receipts are written, and what leaves the machine.

Where Excellent stops

  • Verification evidence does not sync between teammates yet. Receipts stay on the machine that produced them.
  • There is no hosted service, no SSO and no invite flow.

Someone else does this work at your company?

Run one real task and see what held up

curl -fsSL https://excellent.so/install.sh | sh