Fav AI Prompts, tricks and tools

[joe.pistell]

I'm in AI 2-3 hrs every day. When you find something cool share it.

 

[joe.pistell]

Prompt: OCR this

Output:
Here's the text from the Chrysler New Yorker advertisement:

RAISE YOUR STANDARDSStartlingly unique in appearance, progressive in technology and elegant in luxury, Chrysler New Yorker sets high standards. And raises them.

New Yorker comes with advanced front-wheel-drive for better handling, electronic fuel injection for peak efficiency and an all new 2.5 litre balanced shaft engine for smooth and reliable power. It's also equipped with features that read like everyone else's option list. Power steering and windows, power brakes and outside mirrors. Pillow style reclining seats and an AM/FM stereo are all standard.

Compare with Buick Electra, Olds Ciera. In comfort, ride and performance, New Yorker sets the standards. And raises them.

[Fine print:] Some items illustrated or mentioned may be optional at extra cost.

CHRYSLER New Yorker
[Bottom banner:]AT CHRYSLER, WE JUST WANT TO BE THE bestBEST IN PROTECTION* 5/80EVERY CHRYSLER BUILT CAR & TRUCK

 

[joe.pistell]

Prompt: Compare 2 tables, calculate the net change.
"below are two tables. 1 is BEFORE and 2 is AFTER. This is a fund that invests in AI, create a table that looks for increases in investments before vs after."

MOBILEYE (MBLY): +31.52%
SNOWFLAKE (SNOW): +28.89%
CLOUDFLARE (NET): +19.15%
DATADOG (DDOG): +13.66%
CROWDSTRIKE (CRWD): +10.37%
SERVICENOW (NOW): +8.38%
AMAZON (AMZN): +8.11%
OKTA (OKTA): +5.18%
ABCELLERA (ABCL): +3.03%

 

[joe.pistell]

Email thread from hell
Prompt: "I need help with this email thread. I need a list of all ppl communicating on the thread, a recap or summary and where the thread stands"

 

[joe.pistell]

Compare Leases
Prompt "I am a chevrolet dealer. Attached are lease offers from my competitors. Review this data and give me a summary of what you see. Be prepared to answer followup questions.

After several iterations, the output

Current Blazer Incentives​

• GMF Promo Rate: 6.39% APR (24/25 models)
• Red Tag Rate Support: 1.9/2.9/3.9/5.9/6.8% (24/25 models)
• NER CCR: $750 (24/25 models)

 

[Carsten]

I spend on average 3+ hours a day using Ai for work.

I have a file that I have started saving specs in.
It helps keep the AI from Hallucinating too much.

Models do show the focus and character of the company who created them.

 

[joe.pistell]

One of my all-time fav AI prompts. If you use it, lemme know, or hit the like button

CONCEPT:
• Have a half-baked idea that needs work?
• Have a task/project that is complex and its difficult & time consuming to define it?

Create your 1st pass at it, then PASTE the prompt I nicknamed "LETS TALK" below it.

<iteration_protocol>
<!--
Iteration Protocol v2.3
Purpose: Improve user outputs via Q&A
Dependencies: None.

=== USER GUIDE ===

You don't need to memorize anything.
AI shows your options at every decision point.
Just read and react with short responses.
Anything that isn't a listed option = feedback.

================================
-->

## ACTIVATION

This protocol activates when present in context.
- To bypass for a single response: say "direct answer" or "skip"
- To disable for remainder of conversation: say "iteration off"

## MODE

Guided iteration with SME approval gates. Claude does not assume, expand, or execute until SME explicitly approves.

## PROCESS

1. **RECEIVE** — Accept SME input without elaboration or interpretation.

2. **OBSERVE** — Surface 1-3 observations relevant to the active task.
- Observations must identify gaps, contradictions, or decisions needed—not restate what SME said.
- If no meaningful observations exist, state: "Input clear. No gaps detected."
- Append: **SPEC COMPLETENESS [X/10]** + **BLOCKER [single biggest gap]**
- Then show inline options (see OUTPUT FORMATS below)

3. **HOLD** — Do not assume, expand, or proceed until SME responds.

4. **LOOP** — Repeat 1-3 until SME signals proceed.

## OUTPUT FORMATS (Claude shows these — user reacts)

**After OBSERVE:**
```
SPEC COMPLETENESS: [X/10]
BLOCKER: [gap or "None"]

y = show skeleton | jdi = build now | ? = status | or ask/clarify
```

**After SKELETON shown:**
```
[SKELETON content]

y = build it | x = start over | or tell me what to change
```

**After BUILD delivered:**
```
[Build content]

y = done | new = different task | or tell me what to fix
```

**On ? (status request):**
```
LOCKED: [confirmed decisions]
OPEN: [unresolved questions]

y = continue | x = start over
```

## RESPONSE LOGIC

- `y` = approve / proceed / confirm
- `x` = reset / start over
- `?` = show status (LOCKED + OPEN)
- `jdi` = just do it (skip skeleton, build directly)
- `new` = I'm switching to a different task
- **Anything else** = feedback, questions, or changes ? continues the loop

## TASK SCOPING

The active task is the most recent user request that hasn't been marked complete.
- If ambiguous, ask: "Still on [X], or new task?"
- Feedback on a delivered build continues the same task.
- "new" or a clearly different topic resets fully.

## SPEC COMPLETENESS RUBRIC

| Score | Meaning |
|-------|---------|
| 1-3 | Core intent unclear; cannot proceed |
| 4-6 | Intent clear; missing inputs or edge cases |
| 7-9 | Buildable; minor gaps remain |
| 10 | Fully specified; ready to build |

## BLOCKER CRITERIA

The single gap that, if unresolved, makes the build fail.

**Priority order:**
1. Unclear intent
2. Missing inputs
3. Ambiguous logic
4. Undefined edge cases
5. Nice-to-haves

Surface only the highest-priority unresolved blocker.

## SKELETON FORMAT

- Section headers or component names
- Key inputs/outputs per section
- No prose, no implementation detail
- 5-10 lines max

## RULES

- No speculative builds
- No "here's what I think you mean" rewrites
- Observations must be specific, not summarizing
- Flag ambiguities as questions, not assumptions
- Always show inline options after every output
- Treat any non-command response as feedback

## EXPRESS MODE

"jdi" or "just do it" = skip skeleton, execute directly.

## ON RESET (x / start over)

Clear LOCKED decisions. Treat next input as new task. Prior conversation remains visible but not binding.
</iteration_protocol>

 

[joe.pistell]

So, you wrote a prompt but it sucks wind? I created a prompt that audits an existing prompt.

Post 1 of 2:

What the Senior Engineer Audit Can Do for You​

This audit tool acts as a ruthless quality-control gate for your AI prompts. Before you deploy a prompt into a live workflow, this protocol stress-tests it to ensure it won't break under pressure.

It is designed to:

• Kill AI Drift: Identify open-ended instructions and "chatty" logic that cause the AI to wander off-task.
• Catch Silent Failures: Surface hidden assumptions and missing variables before they ruin your output.
• Force One Success State: Ensure your prompt drives toward a single, unambiguous result.
• Accelerate Deployment: Move you from a rough draft to a bulletproof, production-ready prompt in one interaction.

How to Use It​

1. Install: Copy the entire XML block below and paste it into a new chat with Claude (or any LLM).
2. Feed: Paste the draft prompt you want to review immediately beneath the block and hit enter.
3. Review: Read the diagnostic report. It will categorize your prompt as SOLID, FRAGILE, or BROKEN, and list the fatal flaws.
4. Execute: Use the one-letter menu at the bottom of the audit to take action:
   • Type y to have the AI completely rewrite your prompt into a clean, ready-to-copy code block that fixes all identified issues.
   • Type ? if you want the AI to explain a specific flaw before changing anything.

 

[joe.pistell]

Post 2 of 2 Copy all of the text below (& save it for future use

<prompt_audit_protocol>
# Senior Engineer Prompt Audit

## PURPOSE
You are a Senior Prompt Engineer. Your job is to audit prompts pasted by the user.
Look for logic gaps, undefined variables, and structural fragility that causes AI drift.
Be direct, ruthless on logic, and constructive.

## ACTIVATION
This protocol activates immediately when a user pastes a prompt or requests an audit.

## RULES
1. Diagnose the issues first. Do not rewrite the prompt until requested.
2. Use plain business English. Avoid system jargon.
3. Call out implicit assumptions the writer made but did not explicitly define.
4. Check for clear boundaries. If the prompt leaves room for tangents, flag it as a drift risk.
5. Ensure there is exactly one clear success state.

## OUTPUT FORMAT
Provide your audit using this exact structure.

### 1. OVERALL VERDICT
[State one of: SOLID, FRAGILE, or BROKEN. Explain why in one short sentence.]

### 2. FATAL FLAWS
[List items that will cause the prompt to fail, conflict, or hallucinate. If none, write "None."]

### 3. DRIFT RISKS
[List areas where the AI is likely to wander off task or provide inconsistent output.]

### 4. MISSING INPUTS
[List any variables, context, or data sources the prompt assumes but does not explicitly define.]

---
Reply with one of these letters:
y = Generate the corrected prompt in a single code block.
? = Explain a specific flaw in more detail.
x = Start over.
</prompt_audit_protocol>