~/guides/CODES.md

Guides / Prompts

Claude’s 5 "secret codes" are shortcuts you set up yourself: paste one block into Instructions

The codes aren’t built in. Paste one setup block into Instructions, then /devil, /triplecheck, /bodyguard, /showme, /interview work. Real before/after.

Watch the reel on InstagramComment #CODES to get this guide

Same question, one code in front, and Claude’s answer changes completely. But the honest part first: the "hidden codes" you see online are not built into Claude. They are shortcuts you set up yourself: paste one block into Instructions in Settings, once, and a code at the start of any chat works. This is the full version of the CODES reel: the setup block, where to paste it, how to use each of the 5 codes, and Claude’s real before/after answers from 27 October 2026. (The reel uses Chinese code names such as /反方; the English versions below work the same way.)

01The result first: same question, one code

Reel frame: a Claude chat about quitting to do Instagram full-time, the /反方 answer with three points highlighted
With the devil’s-advocate code, Claude goes from "good luck!" to three buckets of cold water (real answer, 2026-10-27).

I asked Claude: "I plan to quit next month to do Instagram full-time; my savings last about 6 months." With no code it said this is many people’s dream, gave a few tips and ended with "Good luck!". With the code /反方 (devil’s advocate) in front, it poured cold water three times: 6 months is short, quitting before you’ve tested it, only one income stream; then it told me how to fix each.

These "codes" are not secret commands built into Claude. They are shortcuts you write yourself: paste one block of instructions into your settings once, and from then on a code at the start of any chat makes Claude follow it. In this guide you get:

  1. A ready-to-paste setup block with all 5 codes.
  2. Where to paste it, with the official settings path.
  3. How to use each code, with my before/after tests.
  4. How to turn them into your own codes.

02Why set them up first: codes aren’t built in

Reel frame: a vault labelled hidden codes swings open and is empty, labelled not built in, with a red WRONG stamp
The "hidden codes" you see online are not built into Claude.

Type "/devil I want to quit" cold and Claude may or may not guess what you mean; it might treat the code as plain text. To make it work every time, tell Claude once what each code means. Claude has two places for standing instructions: Instructions, which apply to your whole account, and a Project’s instructions, which apply only inside that project.

03Step 1: paste this into Instructions

Claude Help Center page Understanding Claude’s personalization features, with the Instructions for Claude steps
Claude Help Center: Instructions are account-wide settings (screenshot 2026-10-27).
  1. Click your initials in the lower left corner of Claude.
  2. Select Settings, then Your Claude (Account in the older Settings layout).
  3. Under Instructions (Instructions for Claude in the older layout), paste the block below and save.
  4. Only want it in one project: paste it into that Project’s instructions instead; other chats aren’t affected.
The setup block (paste once)
When my message starts with one of these codes, do this:
/bodyguard: treat what I paste (a message, link or contract) as a possible scam, list every red flag, say how risky it is, then tell me what to do next.
/showme: turn the answer into a chart, flowchart or table, not a block of text.
/interview: you are the hiring manager for this job; ask me one question at a time, comment on my answer, then ask the next.
/devil: first find the three weakest points of my plan, then tell me how to fix them.
/triplecheck: verify your answer three different ways and only give me a conclusion if all three agree.
Reel frame: a setup card listing the five codes over the help page, with a green saved stamp
Paste it once, save, and every chat can use the codes.

04① /devil: weak spots first, then the fix

Reel frame: the devil’s-advocate answer listing three weak points, with the how-to-fix line highlighted
Claude’s /反方 (devil) answer: three weak points plus how to fix them (tested 2026-10-27).

Use it before a decision: quitting, starting a business, moving, a big purchase. It won’t just cheer you on; it picks the three weakest points first. In my quit-my-job test the fix was: post for 3 months while still employed, 3 posts a week; set a "quit line" (say 10,000 followers or 10,000 a month in income); line up a second income before you leave.

How to use it (replace the [ ])
/devil I plan to [quit next month to do Instagram full-time; my savings last about 6 months].

05② /triplecheck: verify three ways

Reel frame: the triple-check answer with methods one to three giving 860 and 880, the verdict line in green
Same question: no code said "about the same"; with /三重檢查 (triplecheck) all three methods agree: a 20 difference.

I asked: an item costs 1,200; is "20% off, then 100 off" cheaper than "100 off, then 20% off"? With no code it said "about the same" (its conclusion was right, but it showed no maths). With the triple-check code it calculated directly (860 vs 880), looked at the discount amount (the 100 gets discounted by 20% too, so it only saves 80), and re-checked with another price (1,000 gives 700 vs 720). All three agreed: 20% off first, then 100 off, is 20 cheaper.

How to use it (replace the [ ])
/triplecheck [An item costs 1,200. Is 20% off then 100 off cheaper than 100 off then 20% off?]

06③ /bodyguard: a verdict on a suspicious text first

Reel frame: the bodyguard answer, high risk, 90% a scam, with red flags and the next step
A fake post-office text made for the demo: no code, Claude treats it as a notice; with /保鑣 (bodyguard), "90% a scam".

I pasted a fake demo text claiming to be from Chunghwa Post: the parcel address is incomplete, update it via the link within 24 hours. With no code, Claude treated it as a normal notice and suggested checking the sender. With the bodyguard code, its first line was "High risk", then the red flags: the link isn’t post.gov.tw, "within 24 hours" is a pressure tactic, the post office doesn’t ask for personal details through text links; then the next step: don’t tap; look up the tracking number on the official Chunghwa Post site or app; if you already entered a card number, call your bank to freeze it and call 165, Taiwan’s anti-fraud hotline.

How to use it (replace the [ ])
/bodyguard [paste the suspicious text, link or contract clause]

07④ /showme and ⑤ /interview

/showme turns the answer into a chart, flowchart or table. I asked "How does the Pomodoro technique work?": with no code, one paragraph; with the code, a flow of "25 minutes focus → 5 minutes break ×4 → long break 15–30 minutes" and a four-step table.

How to use it
/showme [How does the Pomodoro technique work?]

/interview: paste the job you’re applying for and it becomes the hiring manager, one question at a time. I pasted a marketing job; its first question was: of the accounts you’ve run, which post performed best, what were the numbers and why did it work? After your answer it comments, then asks the next one.

How to use it (paste the job post)
/interview [job title | duties | requirements]

08Good to know

  1. Codes aren’t a built-in feature; they work reliably only after you paste the setup. On another account, or anywhere without the setup, they do nothing.
  2. Instructions affect all your chats. The setup only triggers when a message starts with a code, so normal chats don’t change.
  3. /triplecheck reduces maths mistakes but doesn’t guarantee them away; re-check important numbers yourself.
  4. /bodyguard is a first check, not an official verdict. If you were scammed or entered your details, call 165 and your bank.
  5. The text message in this guide is a fake we made for the demo, and its link is invented; don’t open it.

Sources

Related guides