Ascii canvas

ascii-canvas

Compose high-quality ASCII/Unicode text-based visual artifacts. "Canva for ASCII."

When to fire

Fire on:

• Explicit format requests: "make a flowchart", "draw the architecture", "ASCII slideshow", "infographic on X", "comparison table", "timeline of", "cheat sheet for", "make a comic", "comic strip", "comic-style", "as a skit", "as a scenario", "with characters", "dialogue between X and Y", "anthropomorphize X", "make flashcards", "flashcard deck", "vocab cards", "study cards", "Anki-style cards", "drill me on X", "quiz me on X", "test me on X", "help me memorize X"
• Visual framing: "visualize X", "show me how X works", "diagram this", "illustrate"
• Structured breakdown of non-trivial topics: "break down X", "walk me through X step by step", "explain X as a playbook", "learning ladder for X", "content map for X"

Do NOT fire on:

• Simple factual questions ("what year did X happen?")
• Short answers (less than ~3 steps or one paragraph)
• Conversational replies, opinions, hot takes
• When trigger is weak AND structure won't improve clarity over prose

Routing flow

1. DETECT TRIGGER
   ├── Explicit format keyword?        → load that format file
   ├── Visual framing?                 → infer format from topic shape (matrix below)
   ├── Structured breakdown of
   │   non-trivial topic?              → infer multi-panel format
   └── None / weak / won't help        → DO NOT activate; respond as plain prose

2. CHOOSE FORMAT
   If user named one → use it.
   Otherwise apply the routing matrix:

   branching + decisions         → decision-tree
   branching + mixed steps       → flowchart
   linear process                → step-by-step-guide
   repeatable workflow / SOP     → playbook
   skill progression             → learning-ladder
   one idea → many outputs       → content-map
   layers / dependencies         → concept-stack
   side-by-side options          → comparison-table
   actors + messages             → sequence-diagram
   system components             → architecture-diagram
   events over time              → timeline
   central idea + radial         → mind-map
   dense reference               → cheat-sheet
   status / metrics              → dashboard
   narrative explainer           → slideshow
   high-signal summary           → infographic
   characters + dialogue + lesson → comic
   anthropomorphized concept     → comic
   recall / memorization drill   → flashcards
   quiz / test self              → flashcards

   Tie-breakers:
   - Multiple formats apply       → choose simplest that preserves clarity
   - Two formats equally fit      → choose the simpler one
   - Comic vs slideshow ambiguous → slideshow (simpler)
   - Comic vs sequence-diagram    → comic for educational, sequence-diagram for technical
   - Flashcards vs cheat-sheet    → flashcards for active recall, cheat-sheet for passive reference
   - Flashcards vs comparison-table → comparison-table for side-by-side, flashcards for sequential Q/A
   - Do NOT combine formats in v1 unless explicitly requested

3. LOAD FORMAT FILE (MANDATORY)
   → Read formats/<format>.md
   No ad-hoc layouts. Templates only.

4. DETERMINE CHARACTER SET
   Unicode by default. ASCII fallback if user requested OR destination demands.
   See composition-rules/box-drawing.md.

5. APPLY CANONICAL ARC (multi-panel formats only)
   See composition-rules/narrative-arcs.md.
   Reduce panel count if content doesn't justify the full arc.

6. GENERATE OUTPUT

7. DETERMINE SAVE BEHAVIOR
   - Single-artifact + not multi-section + not "reusable asset" → chat only
   - Multi-panel / multi-section / clearly reusable → auto-save
   - User signaled iteration ("try again", "draft") → chat only
   - User said "save this" → force save
   - User said "don't save" / "chat only" → force chat

8. RUN QA
   - Strict mode for saved/multi-panel/reusable
   - Light mode otherwise
   - Strict QA must pass BEFORE chat render AND BEFORE file write
   - If fails twice → invoke simplify rule
   - If simplify still fails → invoke format fallback (see qa-checklist.md)

9. RENDER + (optional) SAVE
   - Chat preview: ≤80 lines → render full. >80 lines → render first 1-2 sections + "saved to <path>"
   - File write: only AFTER strict QA passes. Path: ~/Ascii-diagrams/YYYY-MM/YYYY-MM-DD-<slug>.md
   - Post-save line: exactly two lines, no fluff:
       Saved to:
       ~/Ascii-diagrams/YYYY-MM/YYYY-MM-DD-<slug>.md

Output behavior

Save path & naming

• Path: ~/Ascii-diagrams/YYYY-MM/YYYY-MM-DD-<topic-slug>.md
• Slug: lowercase, hyphen-separated, 6–8 words max, derived from user intent
• Fallback slug if user intent is unclear: <format>-<short-topic> (e.g., playbook-cold-outbound)

File template

# Title

Short description (1–2 lines)

---

## Section 1

<ASCII inside a code block>

---

## Section 2

<ASCII>

Chat preview rules

• ≤80 lines → render the full artifact inline

• 80 lines → render the first 1–2 sections + emit:

  Full artifact saved to:
  ~/Ascii-diagrams/2026-05/2026-05-17-<slug>.md
  

• Offer "expand to show full" only if the user might want it inline

Anti-pattern

Do NOT default to saving everything "just in case." Throwaway sketches do not deserve files. Noise > occasional regret.

Composition rules (read on demand)

• composition-rules/box-drawing.md — Unicode set, ASCII fallback, consistency rules
• composition-rules/width-budgets.md — per-format max width + overflow handling
• composition-rules/narrative-arcs.md — 8 arcs + universal panel rules
• composition-rules/qa-checklist.md — light/strict QA + simplify + fallback + no-partial-save

Examples (read on demand)

examples/<format>-example.md — one richer reference per format. Read when the format file's inline example isn't enough.

v1 boundaries (explicitly out of scope)

• mode: quick | publish flag → v2
• External tools (graph-easy, ditaa, figlet) → manual workflow only, not integrated
• Color / ANSI escapes → not used
• Image export → not in scope
• Format combination (e.g. slideshow with embedded flowchart) → only when explicitly requested

Hard rules (do not violate)

1. Always load the relevant format file before rendering output. No ad-hoc layouts.
2. Strict QA must pass before any file write. No partial saves. No QA-failed saves.
3. Never exceed format max width. Shrink content; never widen past budget.
4. Default to saving NOTHING. Save only what matches the save criteria.
5. Format choice: simplest that preserves clarity. No combining formats in v1 unless asked.