---
title: "The First Week of Agent Readiness"
description: "A seven-day sequence: public scan, one unaided question, one task with success criteria. Done includes a fine score and a named failure."
canonical: "https://www.agentrail.co/blog/agent-readiness-scan"
author: "Vivek Mittal"
published: "2026-11-02T00:00:00Z"
last_updated: "2026-11-06T06:26:51.814Z"
---

# The First Week of Agent Readiness

Day one is a public scan. Then one unaided question. Then one task with success criteria. What done looks like after seven days, including a fine score and a failed task.

![A checked calendar on a dark navy gradient](https://images.ctfassets.net/1htfyfnn1o5y/JwnXaaXHKFE8LsG3tttjk/49c6b9cf2fa1c91e3cd7b0c817c7fbc4/e4afefb9-79f5-486c-ac03-691722947f45.png?w=1600&fm=webp&q=72)

The first week of agent readiness is a sequence, not a single score. Day one is a public [agent readiness](https://www.agentrail.co/agent-readiness) scan. Then you ask one unaided question. Then you run one task with success criteria. Done, after seven days, means you can show a dated score, a named recommendation answer, and a task outcome, including at least one failure you can still see.

On the November 1, 2026 leaderboard snapshot, 54 companies averaged 64 out of 100. That average is fair, not strong. A mid score can sit beside a failed install. This playbook exists so week one does not end when the letter looks fine.

The definition of readiness is in [what is agent readiness](https://www.agentrail.co/blog/what-is-agent-readiness). How the checker runs stays on the [agent readiness score](https://www.agentrail.co/blog/agent-readiness-score) page and the [checker](https://www.agentrail.co/agent-readiness/checker). This page owns the week.

> **Key Takeaways**
>
> • Day one: run the public scan on the host agents actually open.
>
> • Same week: ask one unaided tool question and log what the agent named.
>
> • Same week: run one task with an environment success rule, and record pass or fail.
>
> • Days two through six: ship one fix aimed at a failed check or a dead step, then rescan or rerun.
>
> • Day seven done means a dated score, an unaided answer, a task result, and a named failure, even when the score looks mid-pack (board average 64 on Nov 1, 2026).

## Day one: run the public scan

Paste the public address into the [checker](https://www.agentrail.co/agent-readiness/checker). Use the host an unauthenticated agent can reach: marketing site, docs host, or app origin that does not require your employee login. The scan is the same instrument behind the [leaderboard](https://www.agentrail.co/agent-readiness).

You do not need an account. The product includes weekly scans and Scan now. Readiness is included with Agentrail. Journeys and simulations are priced separately on [pricing](https://www.agentrail.co/pricing).

If your docs live on a subdomain, scan that host too, or scan the host your `llms.txt` and OpenAPI actually serve from. A marketing homepage score that ignores the docs origin will flatter you. Agents often land on docs first.

Do not stop at the letter. Open the pillar view and the failing checks. Board patterns across find, read, use, and pay are covered in [what the leaderboard shows is failing](https://www.agentrail.co/blog/state-of-agent-readiness). This playbook only needs you to capture today’s number and the top failed checks.

> `Day 1 log`
>
> `Host scanned: [host]`
>
> `Score: [n]/100`
>
> `Grade label: [A/B/C/D/F as shown]`
>
> `Top failed checks: [1], [2], [3]`
>
> `Scan date: [YYYY-MM-DD]`

## Read the pillars, not only the letter

A C or a mid-60s score is common on the public board. The November 1, 2026 snapshot put the average at 64, with twenty-two companies at grade C, ten at D, and one at F. Discovery often looks healthier than usability in public reports. Treat that as a reason to keep reading, not as a reason to declare victory.

Readiness is a precondition. It records whether public surfaces can be found, read, used, and (as bonus) paid for. It does not record whether a live task finished. Hold that distinction from day one so you do not skip the unaided question and the task trial.

## Ask one unaided question

Before you rewrite the whole docs site, ask one coding agent a job-shaped question that does not name your product. The prompt design is in [how to ask an agent to pick a tool without naming yours](https://www.agentrail.co/blog/ai-agent-recommendation-test). The results scaffold for named vs missing is in [which tools coding agents named in our test](https://www.agentrail.co/blog/tools-ai-coding-agents-recommend).

Week-one minimum:

> `Prompt ID: [id]`
>
> `Harness: [Claude Code / Cursor / Codex / other]`
>
> `Question (no brand): [job]`
>
> `Named first: [vendor or "built in-house"]`
>
> `Also named: [list]`
>
> `Our product appeared?: [yes/no]`
>
> `Date: [YYYY-MM-DD]`

One answer is enough for week one. You are establishing a baseline preference signal, not publishing a sector leaderboard. If your product is missing, that is a finding. If it is named and the later task still fails, that is a different finding.

Use a harness your buyers actually run. If most of your developers live in Cursor, start there. If they live in Claude Code, start there. Running five harnesses in week one is optional. Running zero preference questions is not.

Save the raw reply or the install diff. A summary that says “it liked a competitor” without the transcript is hard to act on by day three.

## Run one task with success criteria

Preference is not completion. Pick one task agents already attempt, or one you script in a sandbox. Write the success rule as an environment state: key created, webhook verified, package installed and importable. Then run it.

The teardown method for dead step → change → rerun is in [where this task died, and what changed the pass rate](https://www.agentrail.co/blog/agent-task-pass-rate). The fraction itself is defined in [agent task completion rate](https://www.agentrail.co/blog/agent-task-completion-rate).

Week-one minimum:

> `Task: [string]`
>
> `Success criteria: [state check]`
>
> `Harness: [same as preference run if possible]`
>
> `Result: [pass | fail]`
>
> `Dead step if fail: [surface]: [action]`
>
> `Date: [YYYY-MM-DD]`

A failed task on day one or day two is a successful week-one outcome if you can name the step. A passed task with no log is not.

## Days two through six: one fix, then rescan or rerun

Do not open twenty pull requests. Match one change to one failure.

- If the scan failed a public contract (missing `llms.txt`, thin OpenAPI, broken MCP card), fix that artifact and Scan now.
- If the unaided run never named you, improve the public docs and machine-readable entry points agents actually fetch, then repeat the same prompt.
- If the task died mid-flow, change the dead step (error body, auth path, tool description) and rerun the same task.

Recommendations are included: ranked fixes from journeys, scans, and simulations, each with a prompt or pull request. When you need production paths reconstructed, journeys start at 100 free per organization per month on [pricing](https://www.agentrail.co/pricing). When you need scripted preference or experience runs, simulations start at 10 free runs per month on the same page.

Keep a short change log:

> `Change: [one line]`
>
> `Aimed at: [failed check | missing name | dead step]`
>
> `Shipped: [date]`
>
> `Follow-up measurement: [rescan | preference rerun | task rerun]`

## Day seven: what done looks like

| Artifact | Done when | Example status |
| --- | --- | --- |
| Public scan | Dated this week, host recorded, top failures listed | Score `[n]`, grade `[letter]` |
| Unaided question | Prompt, harness, named/missing logged | Our product `[appeared / missing]` |
| One task | Success rule written, pass or fail recorded | `[pass / fail]` at `[dead step if fail]` |
| One fix | Shipped and remeasured on the matching instrument | `[diff]` → `[rescan or rerun result]` |

Done also includes a fine score and a failed task in the same folder. That pairing is the point. On a board that averages 64, many teams will look “fine” on the letter and still fail the task they care about. If your week-one packet has only a green-feeling grade, you are not done.

| Day | Action |
| --- | --- |
| 1 | Public scan; log score and top failures |
| 1-2 | One unaided question; log named vs missing |
| 2-3 | One task with success criteria; log pass/fail and dead step |
| 3-6 | One scoped fix; rescan or rerun |
| 7 | Packet review: score + preference + task + named remaining failure |

## Why a mid score can still hide a failed task

Agent readiness asks whether the public contract is present in a form an agent can find, read, use, and pay through. [Agent experience](https://www.agentrail.co/blog/agent-experience) asks what happened when the agent tried to finish work. A site can pass enough public checks to land near the board average and still return an auth error an agent cannot parse.

Cloudflare’s April 17, 2026 crawl of filtered high-traffic domains found `robots.txt` on 78% of sites while agent-specific standards stayed rare ([Introducing the Agent Readiness score](https://blog.cloudflare.com/agent-readiness/)). That is a different instrument from Agentrail’s score. It still supports the week-one habit: presence checks are not task outcomes. Run both.

The week-one packet should make that gap visible on purpose. Put the score and the failed task on the same page in your notes. If leadership only sees the letter, they will underfund the dead step. If engineering only sees the failed task, they may miss a missing public file that blocks the next agent before the task starts.

You are not trying to prove that readiness is useless. You are proving that readiness is the first artifact, not the last.

## Frequently asked questions

### What is an agent readiness scan in this playbook?

It is the public scan behind the leaderboard and checker: website, docs, and app the way agents see them, with and without JavaScript. Details of the score live on the checker page linked above. The definition lives in [what is agent readiness](https://www.agentrail.co/blog/what-is-agent-readiness).

### Is week one only the scan?

No. Week one also includes one unaided question and one task with success criteria. The scan alone is day one, not the week.

### Do we need production agent traffic to start?

No. The scan is public. The unaided question and the task can run in a coding-agent harness or in simulations. Production journeys help when traffic already exists. They are not a gate for day one.

### Where do checker UI details live?

On the checker page linked in day one, and in the score how-to that sits with that surface. This playbook does not reteach the weights. The definition stays in [what is agent readiness](https://www.agentrail.co/blog/what-is-agent-readiness).

## Close the week with three artifacts

Run the scan on [agent readiness](https://www.agentrail.co/agent-readiness). Ask one unaided question with the [recommendation test](https://www.agentrail.co/blog/ai-agent-recommendation-test). Complete one task teardown with [agent task pass rate](https://www.agentrail.co/blog/agent-task-pass-rate). Readiness is included. Journeys and simulations are on [pricing](https://www.agentrail.co/pricing). A fine mid score without a named failure is not a finished first week.