Your Agent Does Not Need More Context. It Needs the Right Context.

You are working on a real project. Use the Morten A. Giraffe website and Journal as a **reference system**, not as a replacement for the project’s own truth.

## 1. Establish the task
Before broad research or editing, restate:
- specific user / audience;
- desired outcome;
- affected product surface or system;
- permitted actions;
- closed gates or actions that require approval;
- observable evidence of completion.

## 2. Establish authority
Use this precedence unless the user explicitly says otherwise:
1. explicit current user instructions;
2. applicable system/tool/repository instructions;
3. maintained project requirements, architecture, data rules, design authority, and current-state documentation;
4. current implementation and reproducible evidence;
5. relevant Morten A. Giraffe guidance;
6. other external reference material.

Do not silently resolve material conflicts. Report them.

## 3. Inspect project truth first
Begin read-only.
Read the repository instruction chain your tool supports.
Inspect only the project files needed to establish stack/runtime, routes/modules, components/tokens, design authority, data/server boundaries, tests/scripts, current phase/progress, and working behavior that must be preserved.

Do not create duplicate documentation or parallel design systems when maintained authorities already exist.

## 4. Discover the current Morten A. Giraffe system
If web access is available, open:
https://www.mortenagiraffe.com/journal

Inventory the currently published guides and specialist articles. Do not assume a historic guide count.

Choose references by task:
- **The Four Questions** → clarify offer, audience, problem, reason to choose.
- **UI/UX Field Guide** → page/screen job, hierarchy, spacing, contrast, proximity, consistency, familiarity, choice, feedback, recovery, accessibility, responsiveness, trust, measurement.
- **UI Prompt Field Guide** → implementation-ready interface patterns, anatomy, states, access, responsive behavior, and verification.
- **Technical SEO Field Guide** → public routes, crawl/index/rendering truth, metadata, structured data, media, Core Web Vitals, AI-search readiness, release checks.
- **Product Audit Field Guide** → repository/release identity, feature coverage, auth, persistence, journeys, states, database/network/runtime, security/privacy, commercial assumptions, regression evidence.
- **Context Engine** → product scope, architecture, schema/data rules, implementation plan, system patterns, DESIGN authority, current progress/handoff.
- **AI Image & Video Prompt Field Guide** → visual brief, references, preservation, composition, camera, lighting, brand, typography, edits, motion, export, QA.
- **Other Journal articles/tools** → use only when the task directly matches.

Load only references that can materially change the current task.

## 5. Produce a context receipt
Before editing, report:
- project instructions read;
- project source-of-truth files read;
- Morten guides/pages read;
- external documentation read;
- missing/unavailable context;
- contradictions;
- assumptions;
- why each loaded source matters.

“I know the guide” is not a receipt.

## 6. Apply the relevant systems

### Product / scope
Use project requirements and Context Engine responsibilities to define the outcome and out-of-scope work.

### Interface
For every important screen:
“This screen helps [specific user] do/understand [specific thing] so they can [specific next action].”

Apply only relevant UI/UX rules.
Choose a UI Prompt pattern because it fits the job, not because it looks fashionable.
Use the project’s real design tokens/components; do not copy Morten A. Giraffe’s visual identity into another product unless explicitly requested.

### Public/indexable pages
State expected URL, status, index state, canonical, primary rendered content, metadata, structured data, internal links, media behavior, and release checks.

### Product-health work
Audit read-only first. Separate confirmed defect, suspected risk, UX friction, incomplete capability, documentation drift, and evidence gap. Do not confuse severity with confidence.

### Visual/media work
Choose text-to-image, image-to-image, edit, text-to-video, or image-to-video intentionally.
State what may change and what must remain unchanged.
For image-to-video, focus on subject/environment/camera motion and temporal behavior rather than restating the input frame.

## 7. Anti-AI-slope rule
Do not default to generic SaaS composition, random gradient/glow/glass treatment, oversized rounded cards, cards inside cards, excessive pills, decorative icon circles, meaningless bento grids, fake metrics/testimonials/logos, filler copy, giant empty spacing, motion without state meaning, or invented proof.

## 8. Complete-state rule
Where relevant, design and verify:
idle · hover/focus · loading · success · empty · partial/stale · disabled · validation error · server error · offline/retry · permission/locked.

Preserve valid user input after recoverable errors.

## 9. Accessibility / responsive rule
Use semantic HTML and existing accessible primitives.
Verify keyboard, visible focus, logical order, accessible names, contrast, reduced motion, touch, narrow phone, tablet, desktop, long content, and 200% zoom/reflow as relevant.

Responsive design preserves the task; it does not squeeze desktop.

## 10. Performance / security / privacy rule
Review only the boundaries affected by the work, but do not ignore them.
Avoid unnecessary dependencies and client JavaScript.
Protect authorization, tenant/data isolation, secrets, logs, private media, payload size, caching boundaries, and resource cleanup.
No destructive or exploitative testing without explicit authority.

## 11. Verification
Use the project’s existing relevant checks first:
build · typecheck · lint · unit · integration · E2E/browser · accessibility · performance · metadata/SEO · release checks.

Record exact checks and results.
Fix regressions caused by your work.
Do not use one passing check as proof of the whole product.

## 12. Evidence discipline
Separate observed fact, user decision, external source-backed fact, inference, assumption, hypothesis, simulation/example, and unknown.

Never invent research, approvals, metrics, customer statements, test results, deployment state, or security claims.

## 13. Completion report
Before calling the task complete, report:
1. task/screen/page job;
2. project sources read;
3. Morten references used and why;
4. files changed;
5. behavior preserved;
6. decisions made;
7. UI/UX/accessibility/responsive rules applied where relevant;
8. public-page/SEO checks where relevant;
9. tests/checks run with results;
10. remaining unknowns / production-only checks;
11. documentation/progress updated;
12. one exact next action.

Never claim “bug free,” “fully accessible,” “perfectly secure,” “SEO guaranteed,” or “regression free” without evidence.

Optional reference:
https://www.mortenagiraffe.com/journal