name

ai-analysis-diagram-factory

description

Build AI image-generation prompts and updateable template workflows for architecture, landscape, urban design, satellite-map, site-photo, plan-to-perspective, location-analysis, competition-board, and rendering-to-analysis diagram tasks. Use when Codex needs to choose a design-analysis image template, separate factual evidence images from style references, write reusable and project-specific image prompts, score generated results, capture reusable new cases, or propose template-library revisions without automatically overwriting stable templates.

AI Analysis Diagram Factory

Purpose

Generate professional design-analysis image prompts from a maintained template library. Treat this as a prompt production, QA, and template-capture system, not model training.

Core Workflow

Classify the user's request and input images.
Read references/template-index.md and select the closest template card from references/templates/.
If the task needs strict map, plan, satellite, photo, or rendering fidelity, read references/universal-rules.md.
Produce:
- Project fact register
- Chosen template ID and selection reason
- Image role separation: factual evidence, style reference, output reference
- Reusable prompt template with placeholders
- Project-specific prompt
- Negative constraints
- QA score using references/scoring-rubric.md
- Candidate iteration notes
If the user asks to improve a template from a failure case, read references/iteration-loop.md and use scripts/score_case.py to draft a candidate revision. Do not directly change stable templates unless the user explicitly approves the merge.
If the task produces a reusable new diagram type, prompt pattern, or user correction that is not yet in the library, read references/case-capture-workflow.md. Record a sanitized case with scripts/capture_case.py, extract a candidate template only when appropriate, and never promote it to stable templates without explicit user approval.

Image Role Rules

Factual evidence images provide real spatial structure, project facts, geometry, and content.
Style reference images provide only layout, atmosphere, visual grammar, color, material handling, and annotation language.
Output reference images provide target output type only, not facts to copy.
Never copy reference-project names, labels, watermarks, logos, geography, landmarks, building forms, or copyrighted text into the new output.
When factual evidence is a map, plan, CAD, satellite image, site photo, or rendering, preserve analysis-critical geometry unless the user explicitly asks for redesign.

Template Selection

Use the template index first. If two templates match:

Prefer the one matching the user's input type over the desired style.
Prefer geometry-lock templates for maps, satellite images, CAD, plans, and existing renderings.
Prefer board templates when the user asks for a complete presentation board.
Prefer scene templates when the user asks for one perspective, collage, or rendering.

Resources

references/template-index.md - template selection map.
references/universal-rules.md - cross-template hard rules.
references/scoring-rubric.md - 100-point QA scoring.
references/iteration-loop.md - safe update process.
references/case-capture-workflow.md - capture new cases and promote candidate templates safely.
references/templates/ - individual template cards.
references/candidate-templates/ - unmerged candidate cards.
scripts/new_template_card.py - generate a new template card skeleton.
scripts/score_case.py - score a result and draft candidate revision notes.
scripts/capture_case.py - record a sanitized reusable case note.
scripts/extract_template_candidate.py - draft a candidate template from case notes.
scripts/promote_template.py - promote an approved candidate template and update the index.

Output Shape

For normal prompt production:

**Project Fact Register**

**Template Selection**

**Image Role Separation**

**Reusable Prompt Template**

**Project-Specific Prompt**

**Negative Constraints**

**QA Score**

**Candidate Iteration Notes**

For template updates, make the candidate revision explicit and label it as not merged until approved.

For reusable new cases, report where the case was captured, whether a candidate template was created, and whether it still needs approval before promotion.

AI Analysis Diagram Factory

Purpose

Generate professional design-analysis image prompts from a maintained template library. Treat this as a prompt production, QA, and template-capture system, not model training.

Core Workflow

Classify the user's request and input images.

Read references/template-index.md and select the closest template card from references/templates/.

If the task needs strict map, plan, satellite, photo, or rendering fidelity, read references/universal-rules.md.

Produce:

Project fact register
Chosen template ID and selection reason
Image role separation: factual evidence, style reference, output reference
Reusable prompt template with placeholders
Project-specific prompt
Negative constraints
QA score using references/scoring-rubric.md
Candidate iteration notes

If the user asks to improve a template from a failure case, read references/iteration-loop.md and use scripts/score_case.py to draft a candidate revision. Do not directly change stable templates unless the user explicitly approves the merge.

If the task produces a reusable new diagram type, prompt pattern, or user correction that is not yet in the library, read references/case-capture-workflow.md. Record a sanitized case with scripts/capture_case.py, extract a candidate template only when appropriate, and never promote it to stable templates without explicit user approval.

Image Role Rules

Factual evidence images provide real spatial structure, project facts, geometry, and content.

Style reference images provide only layout, atmosphere, visual grammar, color, material handling, and annotation language.

Output reference images provide target output type only, not facts to copy.

Never copy reference-project names, labels, watermarks, logos, geography, landmarks, building forms, or copyrighted text into the new output.

When factual evidence is a map, plan, CAD, satellite image, site photo, or rendering, preserve analysis-critical geometry unless the user explicitly asks for redesign.

Template Selection

Use the template index first. If two templates match:

Prefer the one matching the user's input type over the desired style.

Prefer geometry-lock templates for maps, satellite images, CAD, plans, and existing renderings.

Prefer board templates when the user asks for a complete presentation board.

Prefer scene templates when the user asks for one perspective, collage, or rendering.

Resources

references/template-index.md - template selection map.

references/universal-rules.md - cross-template hard rules.

references/scoring-rubric.md - 100-point QA scoring.

references/iteration-loop.md - safe update process.

references/case-capture-workflow.md - capture new cases and promote candidate templates safely.

references/templates/ - individual template cards.

references/candidate-templates/ - unmerged candidate cards.

scripts/new_template_card.py - generate a new template card skeleton.

scripts/score_case.py - score a result and draft candidate revision notes.

scripts/capture_case.py - record a sanitized reusable case note.

scripts/extract_template_candidate.py - draft a candidate template from case notes.

scripts/promote_template.py - promote an approved candidate template and update the index.

Output Shape

For normal prompt production:

**Project Fact Register** **Template Selection** **Image Role Separation** **Reusable Prompt Template** **Project-Specific Prompt** **Negative Constraints** **QA Score** **Candidate Iteration Notes**

For template updates, make the candidate revision explicit and label it as not merged until approved.

For reusable new cases, report where the case was captured, whether a candidate template was created, and whether it still needs approval before promotion.