---

name: intent-map-render description: Use during evo's DRIFT_SURFACED phase (per design §6 state machine) to render diagrams from a frozen intent-map + wiring snapshot + api_delta. Pure-Python deterministic — no LLM calls, no timestamps in payloads, byte-identical output on re-run. Emits 4 diagram types capped at HARD-RULE 5's "3 per consultation turn" — D1 Mermaid sequence (per-component runtime), D2 Cytoscape-JSON blast-radius subgraph, D3 Mermaid sankey for version delta, D4 markdown coverage heatmap. Function-level rendering rejected with explicit error.

intent-map-render (v1)

Pure-Python diagram renderer for the EVO agent (S032).

Status: PRODUCTION (S032 WP-5 ship). Design document: /path/to/project/docs/plans/2026-05-13-evo-agent-design.md §4.1 + §8. No LLM, no timestamps, no network — output is byte-identical on re-run.

Diagram inventory (HARD-RULE 5 cap)

ID	Diagram	Tech	When	Output
D1	Intent Sequence	Mermaid sequenceDiagram	Every run, per component	Mermaid markdown block
D2	Blast-Radius Subgraph	Cytoscape-JSON	On-demand during consultation	JSON for visual-companion HTML
D3	Version Delta Sankey	Mermaid sankey-beta	Mode-b when api_delta.breaking_lines non-empty	Mermaid markdown block
D4	Coverage Heat-map	Markdown table	Every run	GFM table

Hard cap: 3 diagrams per consultation turn. Caller passes diagram_types_requested and the skill rejects requests for >3 distinct diagram types in a single invocation.

CLI

python3 ~/.claude/skills/intent-map-render/scripts/run.py \
  --intent-map <path-to-intent-map.yaml> \
  --wiring-snapshot <path-to-snapshot.json> \
  [--api-delta <path-to-api-delta.json>] \
  --emit D1,D4 \
  --output-dir <where-to-write>

Hard rules

1. No LLM calls. Anywhere. This skill is deterministic-only by contract.
2. No timestamps in payloads. Re-runs produce byte-identical outputs.
3. Hard cap of 3 diagrams per call. The --emit flag accepts at most 3 distinct values from {D1, D2, D3, D4}.
4. C4 container + component level only. Function-level diagrams are rejected with EVO_HARD_RULE_5_VIOLATION (exit 2). No "render every call site" support.
5. Two-tier progressive disclosure on D1. When component count > 20, D1 collapses lanes into <details> blocks (markdown-native, GitHub renders the chevron).
6. Cytoscape vendor offline. D2 uses visual-companion/templates/vendor/cytoscape.min.js when available; CDN fallback only if vendor file missing.

What lives in scripts/

• run.py — CLI entry + dispatch
• d1_sequence.py — Mermaid sequenceDiagram emitter
• d2_cytoscape.py — Cytoscape-elements JSON emitter
• d3_sankey.py — Mermaid sankey-beta emitter
• d4_heatmap.py — Markdown table emitter (intent × test crosswalk)
• loader.py — read intent-map.yaml, wiring snapshot, api_delta JSON

What lives in references/

• diagram-conventions.md — Mermaid syntax details + truncation rules
• cytoscape-elements.md — Element shape contract for D2

Determinism guarantee

Given the same inputs, every diagram is byte-identical between runs. Tested by test_d1_byte_identical, test_d2_byte_identical, etc.

How to invoke

# Emit D1 + D4 for an evo intent-map (typical mode-a output)
python3 ~/.claude/skills/intent-map-render/scripts/run.py \
    --intent-map .ledger/evo/runs/<run_id>/intent-map.yaml \
    --wiring-snapshot .wiring/latest.json \
    --emit D1,D4 \
    --output-dir /tmp/evo-diagrams/

# Emit D3 Sankey when version-upgrade has api_delta
python3 ~/.claude/skills/intent-map-render/scripts/run.py \
    --intent-map .ledger/evo/runs/<run_id>/intent-map.yaml \
    --wiring-snapshot .wiring/latest.json \
    --api-delta .ledger/evo/runs/<run_id>/api-delta.json \
    --emit D3 \
    --output-dir /tmp/evo-diagrams/