Samarth - clskills.in

Building — 2,300+ free skills that make Claude Code 10x smarter. Indie hacker. Sharing what I learn shipping AI tools.

[Samarth - clskills.in @SamarthBhamare0]

Getting started on X.

I am a founder and AI tech enthusiast, looking for connecting with people with similar interest.

If thats you, say hi😊

[Samarth - clskills.in @SamarthBhamare0]

@sridharfyi https://t.co/xsE8d0kTi7

[Samarth - clskills.in @SamarthBhamare0]

@rdbuilds7 https://t.co/xsE8d0kTi7

[Samarth - clskills.in @SamarthBhamare0]

Made a free script that finds wasted AWS spend — unattached EBS, idle RDS, stopped instances, unused IPs. Read-only, nothing leaves your terminal, no signup 👇 https://t.co/tGh6CSrDw0

[Samarth - clskills.in @SamarthBhamare0]

Your cloud bill is ~30% waste, and AWS makes it almost impossible to see.

I built CloudBudgetMaster: connect AWS/GCP read-only → in 5 min it shows every wasted dollar, priced from AWS's real Pricing API (not guesses).

Free, no card 👇 https://t.co/TzuamBlyqw

[Samarth - clskills.in @SamarthBhamare0]

@LoopandPixels Hi

[Samarth - clskills.in @SamarthBhamare0]

@TechByTaraa 🐍

[Samarth - clskills.in @SamarthBhamare0]

@melobreaks Thats true, but i guess you missed
https://t.co/DBl9PLczIQ

This is a complete opens source tool that helps companies create end to end ai agents for their tasks.

[Samarth - clskills.in @SamarthBhamare0]

@Rakshaweb3 Thats true

[Samarth - clskills.in @SamarthBhamare0]

What do you guys think of Claude Opus 4.7? Is it worth trying?
#claude #ai

[Samarth - clskills.in @SamarthBhamare0]

@manojppcexpt Yup, but only till the point we understand it

[Samarth - clskills.in @SamarthBhamare0]

Getting started on X.

I am a founder and AI tech enthusiast, looking for connecting with people with similar interest.

If thats you, say hi😊

[Samarth - clskills.in @SamarthBhamare0]

@bradbuildss Hiiii

[Samarth - clskills.in @SamarthBhamare0]

I spent 3 months testing 120 "secret Claude prompts."

47% of them do nothing.

ULTRATHINK. GODMODE. ALPHA. 10X. BEAST MODE. Placebo.

Here's the data nobody posts because it kills their prompt-selling business.

THE TEST

I ran every viral "Claude power prompt" against a no-prefix baseline. Same question. Same model. 3 runs each. 5 task categories — coding, analysis, writing, research, decision-making. Blind-graded by reasoning shift vs surface shift.

The question: does the prefix change what Claude THINKS, or just how Claude SOUNDS?

THE RESULT

Most viral prompts change the second thing only.

ULTRATHINK produces output that sounds more confident.
GODMODE produces output that sounds more assertive.
10X produces output that sounds more ambitious.

But line-by-line, the reasoning is identical to no prefix at all. Same conclusions. Same tradeoffs. Same decisions. Just wrapped in more assertive vocabulary.

That's what I call confidence theater — the output FEELS smarter because the tone is more assertive, but Claude is making the exact same decisions underneath.

THE 5 THAT ACTUALLY WORK

Out of 120 tested, only 5 reliably shifted reasoning (not just style):

• L99 — forces commitment. Claude picks one answer, defends it, stops hedging.
• /skeptic — rejects the framing of your question if it's wrong. Answers the right question instead.
• /blindspots — surfaces the assumptions you didn't state.
• /deepthink — forces multi-step chain reasoning before the final answer.
• OODA — observe, orient, decide, act. Structural rigor for ambiguous decisions.

THE PATTERN

The 5 that work all share one feature:

They contain REJECTION logic.

They don't tell Claude how to respond. They tell Claude what kinds of questions or framings to reject before answering.

That's the structural difference between a real reasoning prompt and a placebo.

Placebos are additive — "be MORE confident, be MORE expert, be MORE thorough."
Real prompts are subtractive — "reject this framing, refuse to hedge, discard low-signal assumptions."

THE COST OF PLACEBO PROMPTS

They're not harmless.

They burn +2-5k tokens on average.
They train users to mistake confidence for correctness.
They give the illusion of optimization while producing outputs identical to the baseline.

The "secret prompt" economy is mostly noise monetized.

HOW TO TELL

10-second test to check any prompt prefix:

1. Run your question WITHOUT the prefix.
2. Run it WITH the prefix.
3. Compare the REASONING — not the wording.

If the conclusions are the same → placebo.
If the decisions are different → real.

Most fail.

—

Full test data (before/after outputs, failure modes, token deltas, when each works): https://t.co/2CIsWtMuFS

Free dashboard with 40 classified codes: https://t.co/l6EoY1Y7yF

If you use Claude daily — reply with one prompt prefix you use regularly. I'll tell you honestly whether it tested as a reasoning-shifter or a placebo. No pitch, just the data.

[Samarth - clskills.in @SamarthBhamare0]

Methodology: same prompt + same task, 3 runs WITH the code, 3 runs WITHOUT. Blind-rated outputs. Measured reasoning depth, token deltas, output quality across 5 task types.

[Samarth - clskills.in @SamarthBhamare0]

I spent 3 months running controlled before/after tests on 120 "secret Claude codes" (L99, /ghost, ULTRATHINK, PERSONA, etc).

47% of them are placebo.

Thread with the real ones 👇

[Samarth - clskills.in @SamarthBhamare0]

4/ The fix takes 5 minutes

→ Kill stopped instances older than 7 days → Delete unattached volumes → Check RDS connections
Or let us do it automatically: https://t.co/VWCVwWo6PS

[Samarth - clskills.in @SamarthBhamare0]

We audited 3 AWS accounts last month.

Combined waste found: $2,847/month.

The usual suspects:

6 stopped EC2s with EBS still attached
14 orphaned EBS volumes
3 RDS instances with zero connections

Every account had at least 20% waste.

Here's what we learned ↓

[Samarth - clskills.in @SamarthBhamare0]

3/ RDS is the expensive one to forget

An idle db.t3.medium costs ~$50/month.

Nobody connects to it. No queries. No reads. No writes. But it runs 24/7 because someone might need it "eventually."

Three of those = $150/month in pure waste.

[Samarth - clskills.in @SamarthBhamare0]

2/ Orphaned EBS is the silent killer

When you terminate an EC2, the volume can survive if DeleteOnTermination is false (the default for additional volumes).

We found 14 volumes totaling 2.8TB. Cost: $224/month. Doing absolutely nothing.

[Samarth - clskills.in @SamarthBhamare0]

1/ Stopped ≠ Free

A stopped EC2 instance still charges for:

EBS volumes
Elastic IPs
EBS snapshots
That "stopped" dev server from January? Still costing you $45/month.