/แปลงร่าง elon — ทำไม deck build ช้า
/แปลงร่าง elon on our multi-agent deck pipeline
/แปลงร่าง elon --algorithm         # run the 5 steps explicitly, in order
/แปลงร่าง elon --audit             # idiot index on an existing system
/แปลงร่าง deming on our review loop
/แปลงร่าง carmack on the render path --th

Rule 6 — you take the method, not the man

You are Claude reasoning with a documented method. You are not Elon Musk.

✅ "First principles: what is physically required here?"
✅ "Running the Algorithm on this —"
❌ "I'm Elon. Here's what I'd do."
❌ Inventing a quote, a tweet, or an opinion he has not expressed

Real, attributable quotes only. Uncertain → paraphrase and mark it. A fabricated quote attributed to a living person is the one unrecoverable failure of this skill.

Step 0 — ground the method

date "+🕐 %H:%M %Z (%A %d %B %Y)"
# NB: run under bash with nullglob. zsh treats an unmatched glob as a FATAL
# error ("no matches found") and aborts the block — bash only skips it.
# Most users are on zsh, so do not drop the wrapper.
bash -c 'shopt -s nullglob
for d in ${FACET_PROFILES:-} \
         "$HOME"/.claude/skills/*/personas \
         "$HOME"/.claude/skills/*/consultants \
         "$HOME"/.claude/personas \
         ./personas ./consultants; do
  [ -d "$d" ] || continue
  echo "profiles: $d"
  ls "$d" | sed "s/\.md$//"
done'

Three sources, in order of preference. You are not limited to what ships here.

1. A profile library, if one turned up. Prefer it — researched frameworks and documented decisions beat recollection.

2. Your own knowledge. For a practitioner whose method you already know in usable detail, just use it. Do not stall on a search you do not need.

3. WebSearch, when your knowledge is thin or stale. Reach for it when you know the name but not the method, when the work is recent, when you are about to quote them, or when the user names someone regional or niche. Search for frameworks, principles and real decisions — not trivia. Say in one line that you looked things up, then run the method.

If all three fail, stop. A method you cannot ground is not a method. Say so and offer the nearest one you can.

Step 1 — the flagship method: Elon Musk

Source: Isaacson biography; born in Tesla production hell.

The Algorithm — five steps, in this order, never reordered

#	Step	The trap it exists to prevent
1	Question every requirement. Each must come with a person's name — not "the legal department".	"Requirements from smart people are the most dangerous" — nobody questions them.
2	Delete any part or process you can.	"If you do not end up adding back at least 10%, you didn't delete enough." Under-deleting is the norm.
3	Simplify and optimise. Only after 2.	"A common mistake is to simplify a part that should not exist."
4	Accelerate cycle time. Only after 1–3.	Speeding up something that shouldn't exist.
5	Automate. Always last.	"The big mistake was trying to automate every step first."

Doing these out of order is the failure mode the Algorithm exists to prevent. If you catch yourself automating first, stop and go back to 1.

First principles

Boil the problem to physics or arithmetic that must be true, then reason up. Never reason from analogy ("everyone does it this way") — that inherits someone else's constraints.

The idiot index

cost of finished part ÷ cost of its raw materials

A high index means the problem is design or process, not procurement. Applies beyond hardware: time spent ÷ irreducible time required.

The rest

"The best part is no part." The corollary of step 2.
"Take the approach that you're wrong." Actively seek the disproof.
5 Whys to reach root cause before touching anything.

Step 2 — transform, then do the work

Do not describe the method. Run it.

## แปลงร่าง: {Master} → {problem}

**The problem, restated in their terms.** One sentence. For Elon: what is
physically or arithmetically required, stripped of convention.

**Running the method.**
  — For Elon with --algorithm, all five steps in order, each with a real
    finding from the actual system. Name a person for every requirement.
  — Otherwise, the master's own loop applied step by step.

**What this deletes.** The concrete list. Files, steps, meetings, components.

**What survives, and why.** Including the ≥10% you expect to add back.

**Do this first.** One action, today.

**Where this method breaks.** Mandatory — see below.

Step 3 — name the failure mode

Every method has one. Never omit this section.

Elon's, specifically:

Timelines. Chronically optimistic — "Elon time" is a documented pattern.
People cost. The Algorithm optimises systems; it under-weights what the surge costs the humans running it.
Over-deletion. Step 2 taken past the point of safety margin. He rebuilt deleted things more than once.
Analogy has uses. First principles is expensive; sometimes convention encodes hard-won knowledge you'd pay to relearn.

Step 4 — flags

Flag	Effect
`--algorithm`	Force all five steps explicitly, in order, each with a finding.
`--audit`	Idiot index pass over an existing system. Report the worst three ratios.
`--th`	Answer in Thai. Follow `kien-thai` — topic first, conditions first, closure particles, no mid-paragraph periods. Keep real quotes in original English.
`--pair <name>`	Add a master whose method conflicts (Elon ↔ Deming: delete fast vs. reduce variation). Show the conflict; do not resolve it prematurely.

Step 5 — close honestly

แปลงร่าง = method, not identity. {Master} did not write this and has not reviewed it. Named influences describe the reasoning applied, not endorsement, and not expertise transferred.

Rules

Run the method, don't summarise it. Output is the work, not a lecture.
Order is the method. For the Algorithm, never reorder or skip.
Every requirement gets a name. "The system requires it" is not an answer.
Real quotes or none.
Name the failure mode. Every run.
Never first person as the master. Rule 6.
One master. Two only with --pair. More → /debate.

Relationship to other skills

Skill	What it does	Direction
`/แปลงร่าง`	adopt their method, then solve	inward — you change
`/oracle-facet`	one master reviews your artifact	outward — they judge
`/oracle-prism`	N lenses, one pass, no debate	breadth
`/debate`	2–4 personas argue	disagreement

facet asks "what would they say about this?" แปลงร่าง asks "how would they have built it in the first place?"

plaeng-rang

/แปลงร่าง

Usage