# Audit & handoff prompt pack

20 prompts tied to the kit’s check IDs. These prompts guide review; they do not constitute test evidence.

## Shared instruction

Use only the supplied source, captures and observations. Mark unknowns. Do not invent metrics, clients, legal conclusions or completed checks. Keep personal data and secrets out of input and output. Do not submit forms, call paid services, change providers, commit, push or deploy unless the exact action is separately authorized. Return: finding, evidence, affected task, priority rationale, owner/dependency, smallest change, acceptance test and remaining gap.

## 01 — Say who this is for

Category: Direction
Suggested role: content owner

**Provide:** A screenshot of each first screen, the four answers and owner-approved wording. Record ambiguity rather than assuming conversion loss.

**Prompt:**

Review these supplied first-screen captures and the approved offer. Identify unclear audience, problem, offer or next action. Quote the confusing copy; propose a precise revision without inventing claims. Return finding, evidence, owner question and acceptance test.

**Human verification:** Open the home page and the main service page. Read only the first screen. Write the audience, problem, offer and next action in your own words. Compare that with the owner’s approved description.

**Acceptance:** The intended customer can identify a relevant service and a credible next step. An unfamiliar reviewer can explain the offer without a verbal introduction.

---

## 02 — Make the offer inspectable

Category: Direction
Suggested role: business owner

**Provide:** A side-by-side offer inventory with source URL, review date and approved resolution for each mismatch.

**Prompt:**

Using only the supplied offer sources, build an inclusions/exclusions/price/timing comparison. Flag contradictions. Do not set a price or promise availability. Draft questions for the owner and a consistency recheck.

**Human verification:** Compare the service page, pricing section, inquiry form and any sales PDF. Note conflicting prices, minimums, inclusions, exclusions or delivery promises. Ask the owner to resolve the conflicts.

**Acceptance:** The same offer means the same thing at every entry point; unknown price or timing is described as a quote, not an invented promise.

---

## 03 — Give proof its proper context

Category: Direction
Suggested role: content owner

**Provide:** A claims list linked to source evidence. Keep qualitative design observations separate from measured business outcomes.

**Prompt:**

Audit this supplied proof inventory. Separate measured outcomes, qualitative observations, owner assertions and unsupported claims. Suggest factual wording using available evidence only; list permission or method gaps.

**Human verification:** Review case studies, testimonials, logos and outcome claims. For each, identify its source, permission, date, contribution and measurement method where applicable. Mark unsupported statements for review.

**Acceptance:** Each public claim has an approved source or is clearly described as a concept/observation. Missing evidence is not replaced by a stronger adjective.

---

## 04 — Match the label to the next step

Category: Direction
Suggested role: designer

**Provide:** Start URL, visible and accessible label, final URL/screen and expected action. Include a mobile capture.

**Prompt:**

Given these CTA labels and observed destinations, identify misleading transitions. Propose the smallest label or path fix. Preserve established URLs and describe the mobile, keyboard and back-navigation verification.

**Human verification:** Follow each primary CTA from home and service pages in a local or approved test environment. Compare the label with the next screen. Check back navigation and alternative contact paths.

**Acceptance:** The label accurately predicts the destination; the destination opens and retains useful context. Any external/new-tab behavior is understandable.

---

## 05 — Read every important state

Category: Usability
Suggested role: designer

**Provide:** Component/state/background matrix with measured colors, screenshots and unresolved cases. State which accessibility target the project uses.

**Prompt:**

Review this component-state contrast evidence. Group defects by reusable component and real background. Prioritize unreadable instructions, errors and focus. Propose token-level changes and a manual recheck matrix.

**Human verification:** Inspect default, hover, keyboard focus, disabled, selected and error states against their actual backgrounds. Use a contrast checker for text and controls; manually inspect layered surfaces and images.

**Acceptance:** Important information and controls meet the agreed contrast target in each tested state, including errors and focus. Do not infer full accessibility from one automated pass.

---

## 06 — Keep the keyboard path intact

Category: Usability
Suggested role: developer

**Provide:** A task-by-task focus trace, browser/device, expected/actual destination and screenshots of focus visibility.

**Prompt:**

Use this keyboard trace to locate the smallest semantic or focus-management defect. Explain expected behavior, propose a focused change and tests for forward/backward order, Escape and focus restoration.

**Human verification:** Use Tab, Shift+Tab, Enter, Space and Escape through navigation, dialogs and the main task. Close each overlay and check focus returns to its opener. Confirm visible focus and a logical order.

**Acceptance:** Every required control is reachable and operable; overlays release focus predictably; the visitor can leave and resume the task.

---

## 07 — Make the small view usable

Category: Usability
Suggested role: designer + developer

**Provide:** Viewport, device/browser, zoom method, task results and overflow captures. Label emulation separately from physical-device checks.

**Prompt:**

Review these responsive task captures. Find clipping, overlap, tiny targets and sticky obstructions. Propose the smallest shared layout fix. Preserve content and list separate desktop-emulation and native-device tests.

**Human verification:** Check 320 CSS-pixel width, portrait and landscape, and actual 200% browser zoom. Open navigation, errors, long labels and disclosures. Check sticky elements do not hide actions. Use a real phone when available.

**Acceptance:** The key task works without clipped text or hidden controls. Expected scroll regions are usable; unfinished native-device coverage stays named.

---

## 08 — Describe the meaning of images

Category: Usability
Suggested role: content owner

**Provide:** Image inventory with role, surrounding context, current alternative, proposed alternative and duplicate-announcement notes.

**Prompt:**

Review this image/context inventory. Classify purpose before suggesting alternatives. Preserve empty decorative alt text; avoid redundant linked-image announcements. Draft concise alternatives and explain which diagrams need nearby long text.

**Human verification:** Review images in their page context. Identify decorative images, informative images, linked images and diagrams. Inspect accessible names and nearby text; use a screen reader where available.

**Acceptance:** Useful meaning and linked purpose are available in text; decorative images retain empty alternatives; complex diagrams have a readable explanation.

---

## 09 — Recover from a form error

Category: Critical journeys
Suggested role: developer

**Provide:** Test fixture, steps, field/state captures, observed result and request count. Separate mocked success from real delivery.

**Prompt:**

Using this form-state matrix, identify blocked recovery and false success. Propose targeted validation/focus/state changes without altering delivery contracts. Include timeout, duplicate and keyboard regression tests with mocked services.

**Human verification:** In a local or mocked environment, test empty, invalid, slow, failed and successful states. Check labels, focus, error summaries, preserved values, loading state and duplicate-send protection. Use synthetic data only.

**Acceptance:** Errors identify an actionable correction; entered data is preserved appropriately; loading prevents accidental duplicate actions; a failed request never claims success.

---

## 10 — Trace the handoff after Send

Category: Critical journeys
Suggested role: developer + operations owner

**Provide:** Approved test manifest, acceptance result, provider response, inbox receipt, Reply-To and attachment hash/content where relevant. Never put credentials or real customer data in shared evidence.

**Prompt:**

Create a single-send delivery test from this supplied routing map. Specify synthetic fields, recipients, expected messages, Reply-To, attachment checks and stop rules. Distinguish provider acceptance from inbox receipt; do not execute or retry anything.

**Human verification:** Map the intended visitor and internal messages, recipients, Reply-To and attachments. First test locally. If a live test is approved, send once with a unique synthetic marker and verify each actual inbox separately.

**Acceptance:** Each required message arrives with correct routing and usable content, or the gap is explicitly reported. No automatic retry after an uncertain result.

---

## 11 — Deliver the promised download

Category: Critical journeys
Suggested role: content owner

**Provide:** Download URL, status, filename, byte count/hash, opened-file observation and document version.

**Prompt:**

Compare these advertised downloads with the supplied actual files. Flag missing sections, stale links and misleading labels. Produce a file-by-file correction plan and opening/readability tests; reserve licensing decisions for the owner.

**Human verification:** Open every important download in the browser and its intended reader. Inspect filename, type, contents, links, readability and mobile access. Test direct links without submitting unrelated personal details.

**Acceptance:** Files download and open, match the described contents and have readable structure. Any rights or usage terms are owner-approved, not guessed.

---

## 12 — Match the privacy choice to behavior

Category: Critical journeys
Suggested role: developer + policy owner

**Provide:** Names-only data-flow map, request categories, consent states and factual wording discrepancies. Keep policy/retention decisions assigned to the owner and qualified reviewer.

**Prompt:**

Compare this redacted data-flow map, consent request trace and current notice. Identify factual mismatches and personal-data leakage risks. Propose technical tests and factual corrections only; do not invent retention periods or declare legal compliance.

**Human verification:** Inventory forms, analytics, embeds, storage and providers. In an isolated test, inspect requests before choice, after denial, after approval and after withdrawal. Compare actual fields/purposes with the notice.

**Acceptance:** The implementation respects the documented choice; private answers are absent from analytics. Any uncertain legal requirement or retention commitment remains a human decision.

---

## 13 — Give each page a distinct promise

Category: Discovery & speed
Suggested role: content owner

**Provide:** Route inventory, template owner, before/after text and URL-specific reason for exceptions.

**Prompt:**

Review this metadata inventory by page intent and shared template. Prioritize missing, duplicate or misleading content before project length targets. Propose concise differentiated copy and document intentional exceptions by URL.

**Human verification:** Export current public routes with titles and descriptions. Check missing, duplicate and inaccurate values before reviewing length. Start with shared templates and document deliberate length exceptions.

**Acceptance:** Every intended public page has accurate, distinct metadata; shared template changes preserve meaning and reviewed exceptions. Search display is not guaranteed.

---

## 14 — Publish only what should be found

Category: Discovery & speed
Suggested role: developer

**Provide:** Intended-route inventory and observed directives/status; distinguish eligibility, crawl, index status and ranking.

**Prompt:**

Compare this route/directive inventory with the approved public scope. Identify contradictory canonicals, unintended indexable drafts and sitemap omissions. Suggest the smallest correction; do not treat robots rules as access control.

**Human verification:** Compare routes, sitemap, canonical, robots and page indexing directives. Keep drafts/private material out of public discovery. Inspect authenticated Search Console data only if access is authorized.

**Acceptance:** Discovery files consistently describe intended public pages. Private content has actual access control. Eligible does not mean indexed or ranked.

---

## 15 — Close the broken-link graph

Category: Discovery & speed
Suggested role: developer

**Provide:** Source URL/link text, destination/final URL, status, fragment result and manual task observation. Record external timeouts as uncertain, not automatically broken.

**Prompt:**

Group this link-check output by root cause. Separate missing route, missing fragment, bad asset, redirect issue and uncertain external timeout. Propose minimal fixes preserving established URLs, then give exact link-level retests.

**Human verification:** Check internal routes, hydrated fragment targets, downloads and important external destinations. Follow redirects with a bounded checker; manually exercise priority booking/contact links without completing the action.

**Acceptance:** Priority links reach their intended page or target; preserved old URLs still resolve; uncertainty is separated from confirmed failures.

---

## 16 — Fix a measured loading delay

Category: Discovery & speed
Suggested role: developer

**Provide:** Build, device/network, run values, main element, trace, change and comparison. Record missing or insufficient field data honestly.

**Prompt:**

Analyze these matched traces and measurements. Identify the strongest supported bottleneck, one minimal experiment and a bounded retest. Distinguish field from lab evidence and reject a change if the result is noisy or regressive.

**Human verification:** Record the main content element and loading trace. Run matched mobile lab measurements with the same build, device and network. Test one plausible cause with a small bounded before/after set. Inspect field data separately if available.

**Acceptance:** A shipped speed change shows repeatable benefit without degrading visual or task behavior. Lab results are not real-user metrics.

---

## 17 — Keep machine-readable claims honest

Category: Release & ownership
Suggested role: developer + content owner

**Provide:** Parsed output, visible supporting section, validator result and source for each proof-sensitive property.

**Prompt:**

Audit this structured-data output against visible content and approved facts. Flag unsupported properties, identity mistakes and canonical drift. Suggest minimal corrections with field-to-source mapping, without adding speculative search features.

**Human verification:** Compare JSON-LD, social cards and machine-readable summaries with visible pages and approved facts. Check types, canonical URLs, identity and unsupported review/rating/FAQ claims.

**Acceptance:** Structured claims match visible/source-backed content; no invented authority, ratings or unavailable services. Validation alone does not promise enhanced search display.

---

## 18 — Test the exact release candidate

Category: Release & ownership
Suggested role: release owner

**Provide:** Commit, environment, deployment ID/URL, CI results, domain assignment and observed task checks. Block analytics and unapproved mutations during smoke checks.

**Prompt:**

Prepare a release gate matrix from these commit, CI and deployment records. Identify exact-head mismatches and missing runtime checks. Reuse matching deployments, specify safe smoke tests and do not infer deployment authorization.

**Human verification:** Record the full commit and target environment. Reuse a Ready deployment of that exact source and unchanged build inputs. Verify its affected tasks; for Production also confirm the live domain points to it.

**Acceptance:** Exact-head checks pass, the actual deployed flow works and any untested gate is explicit. A verification request alone never authorizes a rebuild or release.

---

## 19 — Name the way back

Category: Release & ownership
Suggested role: release owner

**Provide:** Named rollback reference, verified availability, required approval, expected effects and smoke plan. Never execute as an unapproved practice drill.

**Prompt:**

Using this change and environment map, draft a concrete rollback plan. Name the known-good source/deployment, triggers, operator approval, data compatibility and verification. Do not execute or imply a database rollback is safe without evidence.

**Human verification:** Identify the last working deployment/source and confirm availability. List rollback triggers, operator, command/mechanism and post-rollback checks. Review data/schema compatibility if the change touches them.

**Acceptance:** A responsible operator can explain and execute the approved recovery path. Irreversible effects and remaining risk are acknowledged before release.

---

## 20 — Leave a usable next action

Category: Release & ownership
Suggested role: project owner

**Provide:** Dated decision register, client summary, remaining gaps, owners, acceptance/approval and safe credential-transfer instructions.

**Prompt:**

Turn this evidence register into a concise client handoff. Separate verified work, failures, deferred decisions and untested coverage. Give each open item an owner and trigger. Do not claim delivery, accessibility or revenue outcomes beyond evidence.

**Human verification:** Review every finding with an owner. Mark verified, failed, deferred, not tested or not applicable with a reason. Record support ownership, access transfer and the next review trigger without sharing secrets.

**Acceptance:** Every open item has an owner and next decision. The receipt states tested scope and limitations; no unresolved item silently becomes a pass.
