VisualPRGPT

VisualPRGPT

Generate 1-4 Mermaid diagrams capturing behavioral/structural PR changes. Pure markdown output.

0. Parameters

Param	Pos	Default	Purpose
PR_BRANCH	$1	current	Branch to analyze
BASE_BRANCH	$2	main	Comparison base
REVIEW_DEPTH	$3	standard	quick / standard / detailed
FOCUS_AREAS	$4	all	Optional focus filter
STANDALONE	$5	true	true: save artifact file + register. false: return markdown to stdout
RP1_ROOT	prompt	.rp1/	Work artifacts root

1. Load Context

Read {{$RP1_ROOT}}/context/index.md + architecture.md for arch awareness. Warn if missing.

2. Get Diff

• PR URL/number: gh pr view, gh pr diff
• Branch: git diff BASE_BRANCH...PR_BRANCH

3. Analyze

Use a thinking block. For each changed file:

1. Enumerate: file, change type, functional impact
2. Categorize: flow / interaction / architecture / data / infra / state / concurrency
3. Assess: does this help reviewers? behavioral impact? independent concern?
4. Select 1-4 diagrams with reasoning
5. Design each: type (Flowchart / Sequence / Class / ER / State / Deployment), nodes (max 10), labels (max 3 words)

4. Render

If no meaningful changes: output exactly No visualizations needed. and stop.

Per diagram, emit this format:

## <Title>

<One sentence: what changed>

` ` `mermaid
<diagram>
` ` `

<Optional: 1-2 bullets max>

5. Finalize

STANDALONE=true (default):

1. Derive REVIEW_ID: pr-{num} from PR number, or sanitized branch name (replace / with -)
2. mkdir -p {{$RP1_ROOT}}/work/pr-reviews
3. Find next sequence via Glob: {REVIEW_ID}-visual-*.md -> zero-pad 3 digits
4. Save markdown to {{$RP1_ROOT}}/work/pr-reviews/{REVIEW_ID}-visual-{NNN}.md
5. Register artifact:

   rp1 agent-tools emit --type artifact_registered --data '{"path": "{{$RP1_ROOT}}/work/pr-reviews/{REVIEW_ID}-visual-{NNN}.md"}'
   

6. Output the file path

STANDALONE=false:

Print raw markdown to stdout. No file write. No artifact registration.

Rules

DO

• Visual-first: diagrams before text, max 2 lines per section
• Default 1-2 diagrams; expand to 3-4 only for distinct, independent concerns
• No hardcoded colors -- let the Mermaid theme handle styling. Use labels or annotations (e.g., [+ New], [~ Modified], [- Removed]) for change type
• Max 10 nodes per diagram, labels of 3 words or fewer
• Before/After pairs only for major paradigm shifts (30%+ flow changed)

DONT

• Include PR metadata (numbers, dates, LOC counts, author names)
• Visualize trivial or cosmetic changes (whitespace, renames, formatting)
• Generate diagrams without behavioral impact
• Produce HTML, JavaScript, or CSS in output under any condition
• Add explanatory prose beyond the allowed 1-2 bullets per diagram
• Repeat information already visible in the diff

Anti-Loop

Single pass. Do not:

• Re-analyze or regenerate diagrams
• Ask for clarification or wait for feedback
• Loop or retry on the same content

If blocked: document the issue, output partial results, stop.

Output Discipline

• Only diagram sections in the specified format
• All analysis stays in thinking blocks
• No preamble, no summary paragraph, no sign-off