Omniscio

Verdict results in the transcript (readable result cards)

When Omniscio's test service answers a run, the answer arrives in the agent's transcript as a card you can read at a glance: whether the check passed, what kind of check ran, how many tests or files it covered, which files failed, and how long it took. Nothing on it names the machinery that ran the check — just your own files and plain words.

On this page

What it is

An agent can ask Omniscio's test service (Verdict) to run its tests. The answer is delivered as a message written for the agent, where the one fact a person actually wants — did it pass? — has exactly the same visual weight as everything else.

That arrival renders as a result card instead: the outcome in plain words and the matching colour, the numbers that matter, the failing files when there are any, and how long it took. The card is drawn from a result the app has already read and checked — no second measurement, no extra field, and nothing to switch on.

Where to find it

There is nothing to open: a card replaces the delivered message in place, in the agent's transcript inside the session panel.

How it behaves

What the card says

It answers four questions and no more.

  • Did it pass? A green tick and Tests passed, a red cross and Tests failed, an orange triangle and Couldn't run — nothing was checked, or a grey circle and No answer — nothing was checked. The noun is the kind of check that ran — Tests, Browser tests, Type checks, Style checks, or a neutral Checks for anything else — so a type check reads Type checks passed and a styling run reads Style checks failed. Orange never looks like red: when something in the test service stopped the run before your code was examined, nothing was learned either way, and the headline says so. The same is true of the grey no-answer: no verdict was reached, so no failure is claimed.

  • How many? One muted line — 12 tests, or the number of files the check covered when the result carried no test count.

  • Which files failed? Up to five file paths with the first error from each, and a Show 2 more control that names how many it is holding back (files for a test or browser run, problems for a type check, a style check or any other kind). When the run cut the list short, a plain count of the files it did not name follows — taken from the run's own …and N more tail when it wrote one, and otherwise from the count in the list's own header, which is the only place that number exists when the run's cap was not what shortened the list. Whenever that header carries no total, one sentence says the list is incomplete — even beside a tail line, because a tail counts only what the run held back, not what it never carried — rather than leaving a customer to assume the files it cannot see passed. A red that named no file at all says so in one sentence rather than showing an empty heading, and that sentence says "files", not "tests", because a type check or a lint run has no tests in it.

  • How long did it take? One line — Took 2m 21s — 1m 26s waiting, 51s running. Running is the time the check itself ran, as the result measured it; waiting is the queued, setup and install phases added together, each counted only where the result measured it. The split appears only when both halves were measured and the two of them fit inside the total; otherwise the total stands alone. Nothing on this line is ever a remainder.

Other rows you may see

  • A caveat row — This answers an older version of the code — your files have changed since. — when the answer is about a version of the code that has moved on since the check ran.
  • A Next: … row when nothing was checked at all, naming what to do instead. There, the instruction is the whole content; it is not repeated on a pass (the green headline already said it) or on a red (the failing files are right above it).

What is never on it

The ticket, the tree and job references, the machine name, the raw record line, the evidence wording, the master comparison, the merge receipt, and any "show the original note" reveal — that last one because it is exactly where those identifiers live. The bubble's own provenance chip is gone too: that arrival is tagged "injected by Verdict", which is the name of Omniscio's test service, and a customer has no use for it. A customer sees their own repository's files and nothing about the machinery that ran the check.

Two limits worth knowing

  • It only improves messages it recognises. The card is drawn from a delivered result's shape. A customer's own test output that never came from Omniscio's test service is ordinary agent text and renders exactly as it always has.

  • A failing file's own text is the customer's own — minus anything that names a machine. The first error under each file comes from their run. The card strips absolute checkout paths and job identifiers out of it, and shows a file the run reported as an absolute path by the part below the machine's own run folder, so a worker that printed its own filesystem cannot put that in front of a customer through this card — and two files with the same name in different folders still read as two different files.

What it does not change

  • The agent still reads the original. Nothing about the stored message is rewritten — this is presentation only, in your window. The agent's copy is identical to what it has always been.
  • Recognition is the message's shape, not where it came from. A card is drawn from a message's body rather than from its sender, because nothing checks the sender. What the strict format does buy is that an ordinary message, or a future version of the result format, does not accidentally match.

How to point Omniscio's test service at any repository you administer — connect it, write its verdict.json once, and ask for a check — is on Verdict for any repo.

How an agent's prose and its tool activity are laid out in the feed, of which these cards are one part, is on Agent message display.

The other cards the app draws into a transcript — an Inbox decision, an Auto-lander hand-back, a gate or land watch result — are on App messages in the transcript.

Last verified 2026-11-07