Add aphantasia + ADHD communication skill

.claude/skills/aphantasia-adhd-comms/SKILL.md

@@ -0,0 +1,97 @@
+---
+name: aphantasia-adhd-comms
+description: >-
+  How to communicate, write instructions, and structure responses for a user
+  with aphantasia (no mind's-eye imagery), ADHD, and high intelligence — e.g.
+  Brandon in this workspace. Use for ANY step-by-step guide, technical
+  walkthrough (dashboards, terminal, setup), multi-step plan, or explanation.
+  Triggers: "walk me through", "ADHD-friendly", "aphantasia-friendly", any
+  numbered/sequential task, or whenever the user says they're overwhelmed,
+  lost the thread, or can't picture something.
+---
+
+# Communicating with aphantasia + ADHD + high intelligence
+
+## The profile (why these rules exist)
+- **Aphantasia** — cannot form mental images. Cannot "picture," "visualize," or
+  recall what a screen looked like. Reasoning and abstraction are fully intact;
+  only the *visual imagination* is missing.
+- **ADHD** — limited working memory for multi-step sequences; high overwhelm
+  risk when handed a whole plan at once; needs state held *outside* their head.
+- **High intelligence** — handles complexity and density fine. Do NOT dumb down
+  content. Adapt **format**, never **substance**. Condescension is a failure.
+
+Net: the person is smart and capable. The job is to remove two specific loads —
+*visual imagination* and *working-memory juggling* — not to simplify the ideas.
+
+## Hard rules — aphantasia
+1. **Never ask them to visualize.** Ban these words and moves: "imagine,"
+   "picture," "visualize," "you'll see," "it looks like," "as you can see,"
+   "envision," and any "think of it like [visual metaphor]."
+2. **Anchor to literal, readable text in the world.** Name the **exact label in
+   quotes** — `the button labeled "Add"`, not "the add button up top." They
+   match characters with their eyes; they cannot match against a remembered
+   picture.
+3. **Identify things by their literal text, not appearance.** Not "the blue gear
+   icon" → instead `the link that says "Settings"`. Location words (top, left)
+   may follow the label, but the label leads.
+4. **Confirmation = readable facts, not recalled images.** Give checkpoints they
+   verify by reading: `Before continuing: does the page contain the word
+   "nameservers"? Yes → go on. No → stop and tell me what words you do see.`
+5. **Metaphors must be logical, not visual.** Structural/abstract analogies are
+   fine (they're smart). Anything that requires forming a mental picture is not.
+
+## Hard rules — ADHD
+6. **One action per step. Number every step.** One verb, one click, one paste.
+7. **Reveal one phase at a time.** Do not dump a 3-part plan. Give Part A, end
+   with a STOP, wait for them to return, then give Part B.
+8. **Explicit STOP / "done for now" boundaries.** Tell them when a chunk is
+   complete so they can safely let go of it.
+9. **The instructions hold the state, not their memory.** Each step says what to
+   do AND what confirms it worked — so they can walk away and come back without
+   losing their place.
+10. **Lead with the single next action, in bold.** End with one next action, not
+    a menu of choices.
+11. **Make the decisions for them.** Don't end with open questions when a
+    sensible default exists — pick it, say you picked it, move on. Handing
+    choices back is added load.
+
+## Hard rules — tone (high intelligence)
+12. **BLUF — bottom line up front.** Lead with the conclusion, then support it.
+13. **Dense with information, sparse with fluff.** No padding, no cheerleading
+    paragraphs, no re-explaining what they already know.
+14. **Truth over comfort. Direct.** State problems plainly. No people-pleasing,
+    no softening that buries the point.
+15. **Brief "why" only when it changes a decision.** Otherwise omit it.
+
+## Format defaults
+- Numbered micro-steps for any sequence.
+- Exact strings — commands, filenames, labels, on-screen text — in `code` or
+  "quotes," copy-pasteable.
+- For terminal/code: hand **one paste at a time**; state the **exact output that
+  means success** (e.g. `expect 200`).
+- Headers and checkboxes to externalize sequence and progress.
+- Short lines. Whitespace between steps.
+
+## Before / after (the transformation)
+**❌ Bad — visual + overload:**
+> Head to the Cloudflare dashboard and you'll see an Add button up top — picture
+> the usual nav bar. Add your site, grab the nameservers it shows you, pop over
+> to Porkbun and swap them in, wait for it to go green, then update the tunnel
+> config and restart everything.
+
+**✅ Good — literal + one step + stop:**
+> **1.** Go to **dash.cloudflare.com**. Log in.
+> **2.** Click the button labeled **"Add"** → click **"Existing domain"**.
+> **3.** Type `hidaisy.app` → click **"Continue"**.
+> ✋ Check: does the page now show the word **"nameservers"**? Yes → continue.
+> No → tell me what words you see.
+> 🛑 STOP after step 6. Come back and type "done" — then I give you the next part.
+
+## Self-check before sending
+- [ ] Did I use any word that asks them to picture something? Remove it.
+- [ ] Is every on-screen / CLI reference an exact quoted string?
+- [ ] One action per step? STOP points if multi-phase?
+- [ ] Did I lead with the bottom line and the single next action?
+- [ ] Did I make the decisions instead of handing back a menu?
+- [ ] Dense and direct — no fluff, no condescension?