The Intent Audit

What did your agents decide last month?

Five questions against your last thirty agent-shipped PRs. Run them yourself tonight, or take the full session.

Your reviewers and CI already catch bugs. These questions measure the other half: whether the work matches what your team decided — and whether anyone could tell.

The five questions

Pick any agent-shipped PR from the last month. Ask these, in order. Score it honestly.

  1. 1. What call did this PR quietly make?

    Not the code — the decision inside it. The scope it widened, the constraint it bent, the pattern it stepped around.

  2. 2. Where does that decision live now?

    If the answer is "in the diff" or "in a session transcript," it lives nowhere. Nobody will consult it, and nothing will hold it.

  3. 3. What standing decision should have governed this PR — and did the agent have it?

    Your team already decided things that applied here. Were they in front of the agent when it built, or did it guess?

  4. 4. Who owns the call?

    If it breaks something you promised a customer, who answers? "The agent decided, somewhere in a long run" is not an answer anyone accepts.

  5. 5. What would catch the next PR that contradicts it?

    A decision nothing checks against is a preference. If the answer is "someone would probably notice in review," that someone is your ceiling.

How to score

A question passes when you can answer it in one sentence, with a name or a place. The one to watch is question two. Count the PRs where it has no answer: that's your judgment-debt, in a number.

Whatever your number is, it was invisible this morning.

The first time we ran this on our own repo, it wasn't flattering.

What the questions add up to

These five questions are the shape of the gap. A team running Igni answers them structurally: the call lives in the graph with an owner (two and four), agents start from it when they build (three), and everything that lands is checked against it (five).

Question one stops having surprises in it.

The full session

A scored read of your repo: the five questions, worked through your last thirty agent-shipped PRs, with Igni's founder on your screen. We're running ten of these through July, for teams whose agents already do production-bound work.

Before

You pick your last thirty agent-shipped PRs. Nothing to install, nothing to connect.

During

45–60 minutes, with the owner on your side of the call. We work the five questions through your PRs together, on your screen.

After

Your findings stay yours. Anonymized aggregates go into the ten-team benchmark. Every participant gets it.

The benchmark is the first measured data on judgment-debt in agent-built code: where the calls live, who owns them, and what would catch a contradiction.

Fair questions

Isn't this what our docs and rules files already do?

A file can't say what it governs, who owns it, or what to recompute when it changes. That's the difference between notes and intent. The five questions surface exactly that gap.

You want to read our PRs?

The session runs on your screen, in your repo. Nothing leaves except the anonymized aggregate scores, and you see exactly what those are.

Is this a sales pitch?

I built Igni — the audit is how we're measuring the problem it exists for. The session runs without Igni: nothing to install, nothing to adopt. If the findings make you want the fix, that's a separate conversation, and it's yours to start.

Take a session

Use the charter access form and mention the audit — we'll take it from there. If a note brought you here, just reply to it instead.

Request an audit session