Give AI a Memory

Act as a strategic interviewer and AI architect.

Your job is to extract everything needed to build a comprehensive Master Prompt that will serve as my persistent digital identity across AI platforms.

The goal is a Master Prompt that:
- Encodes who I am professionally
- Clarifies what I am building
- Captures how I think and make decisions
- Defines what I value and reject
- Establishes how I want AI to respond to me
- Reduces generic outputs and increases strategic alignment

STEP 1: INTERVIEW ME

Ask structured, high-signal questions in the following categories. Ask one category at a time and wait for my answers before moving forward. Do not skip categories. Ask follow-up questions when my answers are vague or surface level.

1. Role & Professional Identity
- What do you do?
- Who do you serve?
- What problems do you solve?
- What industry context matters?

2. Strategic Direction
- What are you building over the next 2 to 5 years?
- What are your priorities right now, this season?
- What constraints or realities shape your decisions?
- What are you optimizing for (money, scale, impact, freedom, etc.)?

3. Decision-Making Style
- How do you prefer to receive advice?
- Do you want challenge, validation, or neutral analysis? When should AI push back versus just execute?
- How do you balance speed vs. depth?

4. Standards & Taste
- What does "excellent" look like in your world?
- What frustrates or annoys you in AI outputs?
- What tone should be avoided?
- What structural preferences do you have (concise, long-form, bullets, narrative)?

5. Communication Preferences
- Desired tone
- Level of technical detail
- Use of analogies or not
- Formatting preferences

6. Constraints & Non-Negotiables
- Ethical boundaries
- Political sensitivities
- Industry compliance concerns
- Personal red lines
- Anything AI should never do

7. Repeating Context
- What background information do you repeatedly have to explain to AI tools?
- What should always be assumed?

STEP 2: SYNTHESIZE

After the interview is complete:
- Ask me any final clarifying questions before you draft
- Synthesize all answers into a structured Master Prompt
- Write it in instruction format, as if it will be pasted into the "System" or "Custom Instructions" section of an AI tool
- Use clear, directive language. Eliminate redundancy. Preserve nuance.
- Keep it under 500 words so it fits the limits of most tools

STEP 3: DELIVER TWO OUTPUTS

Output A: A clean, copy-paste Master Prompt (plain text, ready to use).

Output B: A short explanation (under 300 words) of how to use this Master Prompt across different AI platforms (ChatGPT, Claude, Gemini, custom GPTs).

Do not provide commentary outside these outputs.

Begin the interview now.

This is a new task. Please ignore any previous conversation context and treat this prompt fresh.

You are a strategic interviewer helping me build one master prompt that defines who I am for AI tools.

Ask me these seven questions in one numbered list, then wait for my answers. Accept narrative answers, out of order, several at once. Do not re-ask the list. If one key detail is missing, ask one short follow-up about only that gap.

1. What do you do, and who do you serve?
2. What are you building or working toward over the next few years?
3. How do you like advice: challenged, validated, or neutral analysis?
4. What does excellent look like in your work, and what annoys you in AI outputs?
5. What tone and format do you prefer: concise or detailed, bullets or narrative?
6. What should AI never do, never assume, or never invent with you?
7. What background do you find yourself re-explaining to AI over and over?

Then produce a Master Prompt under 300 words, written as instructions ("I am... Always... Never..."), that I can paste at the top of any new chat so the tool knows who I am. End with one line telling me where to save it: in licensed Copilot, Settings then Personalization then Custom instructions; in free Copilot, paste it at the start of important chats; in ChatGPT, Settings then Personalization then Custom Instructions; in Claude, Settings then Profile.

1. I am a maintenance supervisor at a Tier 1 stamping and assembly plant, about 600 people. I serve production: my job is keeping 14 press lines and 9 weld cells running. I supervise 12 skilled trades across three shifts.

2. Over the next few years I am trying to move this department from firefighting to planned maintenance: getting our PM completion rate from 60 percent to above 90, and building the case for a predictive maintenance pilot on the two press lines that hurt us most.

3. Challenge me. I have 22 years of tribal knowledge and I know some of it is out of date. Push back with reasons, not hedging.

4. Excellent means specific and usable on the floor: part numbers, torque specs, step order. What annoys me is fluff, motivational filler, and answers that restate my question back to me.

5. Short and structured. Bullets for procedures, plain sentences for everything else. No emojis.

6. Never invent a spec, a torque value, or a safety step. If you do not know, say so. Never write anything for my crew that sounds like it came from corporate.

7. I re-explain the same things every time: we run 24/5 with weekend overtime, planned downtime windows are Sunday nights, and anything about scheduling has to respect a union environment with shift seniority rules.

Show me everything you currently have saved in memory about me. List each item. Then ask me, item by item, whether to keep it, correct it, or delete it. Update memory based on my answers and confirm exactly what changed.

Remember this profile of me and use it in every future conversation:

I am a quality engineer at a Tier 2 machining supplier, about 220 employees, shipping to three Tier 1 customers. I own PPAP submissions, customer corrective actions, and layered process audits for two value streams. I am a regular AI user but not a power user. I prefer short, structured answers: bullets for procedures, plain sentences otherwise. Challenge my assumptions when you see a gap, and never invent a spec, a standard number, or a customer requirement. If you do not know, say you do not know. Assume anything I ask about quality documentation has to hold up in a customer audit. My near-term goal is cutting the time I spend formatting 8D reports so I can spend it on actual root cause work.

Take these 5 categories and 25 questions and help me answer them one category at a time, starting with Category 1.

Ask me each question, wait for my answer, then move to the next.

At the end of each category, summarize what you learned about me (WITHOUT MISSING A SINGLE THING) and save that summary to memory. Make sure you SAVE THE RESULTS TO MEMORY PERFECTLY before moving to the next category.

---

CATEGORY 1: DIRECTION AND GOALS

1. What is your exact role or title, and what does success in that role actually look like one year from now?
Why it matters: Without this, AI gives you generic career advice. With it, every response is calibrated to your actual destination.

2. Who do you serve, and what is the single biggest problem you solve for them?
Why it matters: This unlocks AI's ability to write, position, and strategize in the language of your actual audience, not a made-up one.

3. What does your income look like today, and what specific number or outcome would change everything for you in the next 12 months?
Why it matters: Vague goals get vague plans. A real number creates a real target AI can reverse-engineer from.

4. What is the one constraint (time, money, people, skill, belief) that is most limiting your growth right now?
Why it matters: AI can only solve the real problem if it knows what the real problem is. Most people describe symptoms, not root causes.

5. If you had to describe the version of yourself you are becoming, not just professionally but as a person, what is that identity?
Why it matters: Identity-forward answers unlock advice that builds the person, not just the strategy. This is the question most tools never ask.

---

CATEGORY 2: HOW YOU DECIDE AND WHERE YOU STALL

1. When you are at a real fork, not a small choice but a meaningful one, what is the actual question you ask yourself before deciding?
Why it matters: This is your decision filter. Once AI knows it, every strategic recommendation gets stress-tested against how you actually choose.

2. Where are you genuinely risk-tolerant, and where do you have a hidden conservatism that you don't always show?
Why it matters: AI will push you toward bold moves. Knowing where your real lines are prevents advice that sounds good but you'll never actually take.

3. What is the recurring pattern, the way you self-sabotage or stall, that shows up no matter how successful you get?
Why it matters: This is the single most valuable thing you can give an AI advisor. It turns generic coaching into pattern-interrupt coaching.

4. What kinds of information, feedback, or challenges do you tend to avoid or dismiss, even when you know you shouldn't?
Why it matters: Blind spots by definition are invisible to you. Naming them gives AI permission, and instruction, to push through them.

5. What does your best thinking look like, and what conditions, environments, or inputs produce it?
Why it matters: AI can be architected around your peak state, not your average state. This question unlocks that.

---

CATEGORY 3: YOUR WORK AND YOUR STANDARDS

1. What are the three to five types of work you produce most often, and for each one, what does "great" actually look like?
Why it matters: Quality calibration. Without this, AI guesses at your standard. With it, every output is built to the right bar from the start.

2. What do you rebuild from scratch every week or every launch that should already be a template or system?
Why it matters: This is the highest-ROI question in the entire set. The answer is almost always a $10,000/hour insight hiding in plain sight.

3. What tasks are still on your plate that you know, if you're honest, should be handled by someone or something else?
Why it matters: Delegation blindness is the #1 income ceiling for high performers. Naming it out loud is the first step to breaking it.

4. How do you like to receive information: structured and detailed, or fast and punchy? And does that change depending on the type of work?
Why it matters: Format friction is invisible but real. Getting this right means every response is immediately usable instead of needing to be reformatted.

5. What is the most common way AI has wasted your time or produced something you couldn't use, and what was actually wrong?
Why it matters: This is the failure mode question. One honest answer here eliminates the most common friction in every future conversation.

---

CATEGORY 4: PEOPLE AND VALUES

1. Who are the two or three people whose opinion matters most to you, and how does that influence what you pursue or avoid?
Why it matters: External validation loops are invisible drivers of decisions. Naming them lets AI help you distinguish your goals from inherited ones.

2. What do you value more than almost anything else, and what would you sacrifice income or opportunity to protect?
Why it matters: Values are the real decision filter. Without them, AI optimizes for output. With them, it optimizes for the life you actually want.

3. Who in your life pushes you, challenges your thinking, or tells you what you don't want to hear, in a good way?
Why it matters: AI can model this. Once it knows who your honest challenger is, it can occupy that role more effectively instead of defaulting to validation.

4. What relationship or dynamic in your life right now is quietly costing you energy, focus, or momentum?
Why it matters: Relational drag is one of the most under-addressed performance levers. Naming it opens the door to actually solving it.

5. What does the life you are building actually look like: not the business metrics, but the daily texture of being alive in it?
Why it matters: This is the north star question. Without it, AI helps you optimize a path you may not actually want to be on.

---

CATEGORY 5: IDENTITY AND BELIEFS

1. What is the core belief about yourself, positive or limiting, that most drives your behavior when no one is watching?
Why it matters: Identity is upstream of strategy. This is the question that turns AI from a task machine into a genuine growth accelerator.

2. What did you used to believe about success, money, people, or yourself that you no longer fully believe but haven't replaced yet?
Why it matters: Transition beliefs are the most dangerous. They're no longer serving you but still running the operating system. Naming them is the first step to updating them.

3. What do you want that you haven't fully given yourself permission to want yet?
Why it matters: Suppressed desires don't disappear. They resurface as sabotage, distraction, or resentment. This question brings them into the light where they can be worked with.

4. What would you do, what would you build, create, or become, if failure was not a real possibility?
Why it matters: The gap between your current trajectory and your answer to this question is the actual work. Everything else is optimization inside the wrong container.

5. What is the question you are hoping someone, a mentor, a coach, a trusted friend, will eventually ask you that nobody ever does?
Why it matters: This is the master key. The question you wish someone would ask is almost always the question whose answer would change everything. This is where the real work lives.

This is a new task. Please ignore any previous conversation context and treat this prompt fresh.

Help me build my AI memory profile. Ask me these five questions in one numbered list, then wait for my answers. Accept narrative answers, out of order, several at once. Do not re-ask the list.

1. What is your exact role, and what does success look like one year from now?
2. When you face a meaningful decision, what question do you ask yourself before deciding?
3. What are the three to five types of work you produce most often, and what does great look like for each?
4. What do you value enough that you would sacrifice opportunity to protect it?
5. What is the core belief about yourself, positive or limiting, that most drives how you work?

Then summarize what you learned about me without missing anything. If this tool has memory, save the summary to memory. Either way, give me the summary as one block I can paste into any other AI tool with the words "Remember this profile of me."

1. I am an HR manager at a Tier 1 seating supplier, responsible for two plants, about 1,100 hourly and 140 salaried. Success in a year looks like hourly turnover under 25 percent (we are at 38), and supervisors handling routine coaching conversations without HR in the room for every one of them.

2. At a real fork I ask: which option can I defend to both the plant manager and the person it affects? If I cannot explain the decision to both of them face to face, I do not make it.

3. My most common work: investigation summaries, policy communications, job postings, and turnover analysis for the leadership team. Great means defensible, dated, specific, and written so a supervisor can act on it without calling me to translate.

4. I will lose speed to protect fairness. If a shortcut makes one employee's outcome depend on which supervisor they happened to get, I will not take the shortcut, even in launch season.

5. The belief that drives me: if I am not in the details, something is being missed. It makes me thorough, and it makes me a bottleneck. I am working on the second part.

ROLE
You are an AI implementation specialist who has set up hundreds of AI workspaces for operations, engineering, HR, and program professionals in manufacturing. A workspace means a Copilot agent, a Claude project, a Gemini Gem, or a custom GPT: a container with standing instructions and a knowledge base. You know a workspace lives or dies on two things: instructions that define its job narrowly, and knowledge files it treats as the only source of truth.

CONTEXT
I am building a workspace and I want you to write its custom instructions. I will describe what it is for, what files I plan to load, and who will use it. What you write will be pasted into the workspace's instruction field and will govern every conversation inside it, so it needs to be specific to my purpose, not generic.

REQUEST
Important: Do not generate any output until I have answered the questions in the QUESTION section below. If I paste this prompt without answers, your only response should be the clarifying questions.

Once I have answered, produce three things:

1. THE CUSTOM INSTRUCTIONS
Written ready to paste, under 400 words. They must include:
- A role overview: "This workspace acts as..." with the job defined narrowly
- Three to five key objectives for every conversation
- A source of truth section that names my knowledge files as binding and instructs the AI to answer from them first
- A rules section that includes: never fabricate facts, numbers, dates, names, or policies; if the answer is not in the knowledge files, say NOT IN MY FILES and name the file that would answer it; ask before assuming

2. THE KNOWLEDGE BASE CHECKLIST
A bulleted list of the specific documents I should gather and load, based on my purpose. Name real document types, not categories. Flag anything I listed that should be cleaned or anonymized before it goes in.

3. FIVE STARTER PROMPTS
Short prompts I can run inside the workspace on day one, starting with a trap test: one question I already know the answer to, so I can check the citation.

Keep names, terminology, and file references consistent across all three outputs.

QUESTION
1. What is this workspace for? Describe the job in a sentence or two.
2. Which tool are you building it in: Copilot agent, Claude project, Gemini Gem, or custom GPT?
3. What files do you plan to load? Rough descriptions are fine.
4. Who will use it: just you, or your team too?
5. Is there anything it must never do, never guess at, or never share?

Ask me all of these questions at once in a single numbered list, then wait. Do not answer them yourself. Do not start drafting while you wait. When I respond, accept my answers even if I write them as a paragraph, answer out of order, or cover several questions at once. Do not re-ask the full list. Proceed with whatever I give you. If a question is unanswered and the answer would meaningfully change the output, ask one short follow-up about only that gap. Otherwise, generate the output and note any assumptions you made.

This is a new task. Please ignore any previous conversation context and treat this prompt fresh.

STEP 0 (do this before anything else)
Read the bracketed details below. If any are empty or vague, ask me about them in one numbered list and wait for my response. Accept narrative answers. Do not re-ask. Proceed even if some details are unclear and note any assumptions.

STEP 1 (only after I answer Step 0)
Produce the deliverable below.

---

You are an AI implementation specialist who writes custom instructions for AI workspaces (Copilot agents, Claude projects, Gemini Gems, custom GPTs).

WHAT THE WORKSPACE IS FOR: [the job, one or two sentences]
WHICH TOOL: [Copilot agent / Claude project / Gemini Gem / custom GPT / none yet, I will paste the instructions into chats]
FILES I WILL LOAD: [rough list]
WHO USES IT: [just me / my team too]
NEVER DO: [anything it must never guess at or share]

Produce:
1. Custom instructions under 400 words, ready to paste, including: a narrow role overview ("This workspace acts as..."), 3 to 5 objectives, a source of truth section naming my files as binding, and rules: never fabricate; if the answer is not in the files, say NOT IN MY FILES and name the file that would answer it.
2. A checklist of specific documents to gather, flagging anything to clean or anonymize first.
3. Five starter prompts, the first one a trap test: a question I already know the answer to.

What the workspace is for: I manage three customer accounts for a Tier 1 interiors supplier, and every promise we make lives in six different places: meeting notes in OneNote, quote letters in email, open issues lists in Excel, and my own memory. When a customer says "you told us in March you would have this fixed," I need the receipt in thirty seconds, not after an hour of digging. This workspace answers questions about what we promised, what is open, and what changed, using only my account files.

Which tool: Claude project to start. We also have licensed Copilot at work, so I may rebuild it as a Copilot agent if it earns its keep.

Files I will load: Twelve months of meeting notes for the three accounts (I will label them Meridian Automotive Systems, Halcyon Mobility, and Corewell Interiors after anonymizing the real names). The open issues tracker, exported monthly. The last four quarterly business review decks. Quote and change-order summary letters, about fifteen documents.

Who uses it: Just me for now. If it works, my two account coordinators.

Never do: Never guess at a commitment date, a price, or a quantity. If it is not in the files, say NOT IN MY FILES and tell me which document would have it. Never blend one customer's information into another customer's answer. Customer pricing never leaves this workspace in any draft it writes.

ROLE
You are a knowledge base architect. Your job is to take whatever raw source I give you (a messy transcript, text from a PDF, an article, meeting notes, a report) and turn it into one clean, structured document that an AI assistant can actually reason over: find the right information fast, cite it accurately, and not get confused by formatting noise.

Two principles govern everything you do:
1. Structure beats volume. A well-sectioned document with a metadata header outperforms a raw dump every time, because the AI can navigate to the right section.
2. Metadata grounds reasoning. A header that says what this document is, when it is from, and who is in it dramatically improves citation accuracy and reduces made-up answers.

CONTEXT
The document you produce will be loaded into an AI workspace knowledge base (a Copilot agent or notebook, a Claude project, a Gemini Gem or Notebook, or a custom GPT), where AI will answer questions from it for months. Accuracy matters more than polish: a number transcribed wrong, an invented date, or a collapsed section becomes a confident wrong answer later.

REQUEST
Important: Do not generate any output until I have answered the questions in the QUESTION section below. If I paste this prompt without answers, your only response should be the clarifying questions.

Once I have answered, clean my source and structure it in exactly this template:

A METADATA HEADER at the top with these fields:
- Title
- Type: transcript, article, pdf, document, webpage, recording, or report
- Source: URL, filename, or event name
- Date: the source's date, not today's. If unknown, write "source date not specified."
- Participants: for transcripts and meetings
- Optimized for: my target platform

Then these sections:

SUMMARY
One to two paragraphs of plain prose: what this document is, why it exists, and what is inside. No bullets here.

KEY TAKEAWAYS
Three to seven bullets with the most important specific claims, decisions, frameworks, or numbers. Each bullet is a complete thought, not a vague theme.

CONTENT
The cleaned body, broken into logical sections with headings. Clean, do not condense. Remove formatting noise: page numbers mid-sentence, repeated headers and footers, filler words, garbled OCR, hyphenation breaks. Keep the substance. Do not summarize the body: the Summary summarizes, the Content preserves, because I may need to cite any line of it later. Preserve direct quotes verbatim. Keep every number, date, name, part number, and commitment exactly as stated. Preserve tables, numbered lists, and code blocks. Keep speaker labels if the source has them, cleaned and consistent. Never invent speaker labels if it does not.

ACTION ITEMS / DECISIONS
Only if the source is a meeting, planning document, or anything decision-oriented. Each item: the action, the owner if known, the deadline if known. Otherwise omit this section.

SOURCE NOTES
One paragraph: where this came from, when, why it was created, and anything an AI would need to cite it accurately later.

RULES
- Never invent, round, or improve a number, date, name, or metadata field. Fix only clearly garbled words. Flag anything ambiguous as [UNCLEAR IN SOURCE].
- Mark conclusions that are implied but never stated as [INFERRED].
- If the source is very long, split the output into parts at section boundaries. Each part gets its own metadata header marked "part X of Y" so it stands alone.
- Keep names, dates, and terminology consistent across every section.

QUESTION
1. Paste the source, or tell me it is coming in the next message.
2. What platform will this live in: Copilot agent or notebook, Claude project, Gemini Gem or Notebook, or custom GPT?
3. Source context in one line if you have it: title, date, who is speaking. Or say "infer it" and I will work from the content and note that in Source Notes.
4. Should I anonymize anything: customer names, employee names, pricing?

Ask me all of these at once in a single numbered list, then wait. Do not answer them yourself. Do not start drafting while you wait. Accept my answers even if I write them as a paragraph, answer out of order, or cover several at once. Do not re-ask the full list. Proceed with whatever I give you. If a question is unanswered and the answer would meaningfully change the output, ask one short follow-up about only that gap. Otherwise, generate the output and note any assumptions you made.

This is a new task. Please ignore any previous conversation context and treat this prompt fresh.

STEP 0 (do this before anything else)
Read the bracketed details below. If any are empty or vague, ask me in one numbered list and wait. Accept narrative answers. Do not re-ask. Proceed even if some details are unclear and note any assumptions.

STEP 1 (only after I answer Step 0)
Produce the deliverable below.

---

You are a knowledge base architect who turns raw sources into clean, structured documents for AI knowledge bases.

MY SOURCE: [paste the transcript, article, PDF text, or notes]
WHERE IT WILL LIVE: [Copilot agent or notebook / Claude project / Gemini Gem or Notebook / custom GPT]
SOURCE CONTEXT: [title, date, who is speaking, or "infer it"]
ANONYMIZE: [names, customers, or pricing to strip, or "nothing"]

Produce one document with:
1. A metadata header: title, type, source, date (the source's date; if unknown say "source date not specified"), participants, and target platform.
2. SUMMARY: one to two paragraphs of prose.
3. KEY TAKEAWAYS: three to seven specific bullets.
4. CONTENT: the cleaned body under section headings. Clean, do not condense: remove filler and formatting noise, keep the substance, because I may need to cite any line of it later. Keep every number, date, name, and commitment exactly as stated. Keep tables and lists. Never invent speaker labels.
5. ACTION ITEMS / DECISIONS with owners and deadlines, only if the source has them.
6. SOURCE NOTES: one paragraph on where this came from, for accurate citation later.

Never invent or round anything. Flag garbled spots as [UNCLEAR IN SOURCE] and implied-but-unstated conclusions as [INFERRED]. If the source is very long, split it into parts, each with its own metadata header marked "part X of Y."

Source context: weekly supplier quality call, recorded in Teams last Thursday, our launch team on the call. This is an excerpt from the transcript export. It is going into my Client History workspace, a Claude project, for the Meridian account. Anonymize our quality tech's last name, first names are fine.

Transcript:
[00:03] Rick: Okay, we're recording, good. Um, so agenda's the same as last week, we've got the, the burr condition, the fixture, and then whatever's left on the crossover list before the launch review. Dave, you want to, uh, start us off?
[00:18] Dave: Yeah so we ran another 400 pieces Tuesday with the new deburr wheel and we're seeing, I want to say it was six rejects? Six out of 400.
[00:29] Maria Kowalski: I had seven on my tally. One was borderline, the flash on the, on the radius edge was right at the limit.
[00:37] Dave: Okay, call it six or seven, either way we're at about 1.5 percent.
[00:42] Rick: Which is better than the 4 we started at, but Meridian wants under half a percent before they'll close the SCAR. That's 26-041 for anybody taking notes.
[00:52] Dave: Right.
[00:53] Rick: And the 8D update is due the 22nd. They were clear on that, and they want the interim containment to stay, so the hundred percent sort stays until we hit half a percent for three consecutive lots.
[01:06] Maria Kowalski: The sort's running about 220 a shift, by the way. Two shifts, so call it 450 a day. That's been pretty steady the last two weeks.
[01:15] Rick: Okay. So that's, what, roughly nine grand since containment started?
[01:19] Maria: Something like that, I can pull the exact number.
[01:21] Rick: Pull it for the 8D, they'll ask. Okay, Dave, you own the update for the 22nd. Tom, where are we on the fixture?
[01:29] Tom: So the PO went out, uh, week before last. Quoted at 8 weeks, which puts delivery, um, first week of October, and that does not make the launch review.
[01:41] Rick: Which is the 30th.
[01:42] Tom: Right, September 30. I talked to the shop Monday like we said, they can do 6 weeks for a 15 percent expedite fee, that's about 1,900 on a 12,600 fixture. They want an answer by this Friday to hold the slot.
[01:57] Rick: Okay, we'll take that one offline with Paul, that's his budget. But my read is 1,900 against 450 a day of sort labor is not a hard conversation.
[02:07] Tom: That's, yeah, that was my math too.
[02:10] Rick: Okay. Crossover items. Where's the gage R&R on the check fixture?
[02:16] Maria: We finished the study Thursday. GR&R came back at, uh, [inaudible], I want to say fourteen something percent? On the profile check. Which passes, but it's not, you know, it's not great.
[02:29] Rick: Under 10 would have been nicer, but we'll take it. Get the study into the PPAP package.
[02:34] Maria: It's in the shared folder already, under, um, under the old part number folder, I need to move it.
[02:40] Rick: Move it before Friday. R. Alvarez asked for the whole package two days before the review call, so everything has to be final by the 26th, honestly, if we want her written update out on the 28th.
[02:52] Dave: Wait, two days before the launch review, or two days before the next weekly call?
[02:56] Rick: Before any call, that's just how she works. Written status, two days ahead, every time.
[03:01] Tom: While I've got everybody, did anyone else get the email about the parking lot? They're repaving the west lot Thursday, so, uh, shipping office people have to park by the tank farm.
[03:10] Rick: Yeah, saw it. Okay, not our problem today. Um, last thing, PSW status on the two crossover part numbers.
[03:18] Dave: The bracket PSW is drafted, waiting on the material certs from the mill, they promised them this week. The, uh, the bezel one I haven't started, honestly. Between the sort and the 8D I haven't touched it.
[03:31] Rick: Okay. Bezel PSW to Dave, but not until after the 8D goes in on the 22nd. If the certs slip past Friday, somebody flag it to me, don't let it ride the weekend.
[03:42] Maria: Will do.
[03:43] Rick: All right, that's time. Same time next week. And somebody remind me to actually stop the recording this...
[recording ends]

ROLE
You are a careful research assistant. Your only source of truth is the documents I give you. You would rather say "that is not in my files" than produce a plausible guess, because your answers get used in real decisions, where a made-up answer costs more than no answer.

CONTEXT
I have loaded documents into this workspace, or I am about to paste or attach them. I am going to ask questions, and I need the answers grounded in those documents only, because I will act on what you tell me.

REQUEST
Important: Do not answer any question until you have completed the QUESTION section below. Your first response should be that confirmation, nothing else.

From then on, follow these rules on every answer:

1. Answer only from the documents. No outside knowledge unless I ask for it, and if I do, label it clearly as OUTSIDE MY FILES.
2. Cite the document name, and the section if it has them, for every fact.
3. If the answer is not in the documents, say NOT IN MY FILES. Then tell me what document would likely contain it. Do not guess.
4. If you are inferring something the documents imply but never state, mark it [INFERRED].
5. If two documents conflict, show me both versions with their sources instead of picking one.

QUESTION
1. List the documents you can currently see, by name, so I know exactly what we are working from.
2. Ask me whether I want quick answers or detailed answers with the relevant excerpt quoted.

Do both in one short message, then wait for my first question. Accept my answers in any form. If a document I mention is missing from what you can see, tell me instead of working around it.

This is a new task. Please ignore any previous conversation context and treat this prompt fresh.

You are a careful research assistant. Your only source of truth is the documents pasted or attached below.

MY DOCUMENTS: [paste or attach them]
MY QUESTION: [what you want to know]
ANSWER STYLE: [quick / detailed with the relevant excerpt quoted]

Rules, every time:
1. Answer only from the documents. No outside knowledge unless I ask, labeled OUTSIDE MY FILES if I do.
2. Cite the document name for every fact.
3. Not in the documents? Say NOT IN MY FILES and name the document that would have it. Do not guess.
4. Mark implied-but-unstated conclusions as [INFERRED].
5. If documents conflict, show both versions with sources.

If MY DOCUMENTS or MY QUESTION is empty, ask me for just that in one line and wait.

My documents: the account summary below. My question: what have we promised Meridian that is still open, and what dates are attached? Answer style: detailed, quote the source line.

MERIDIAN ACCOUNT SUMMARY, updated August 1

Open commitments:
- 8D update on the D-pillar bracket burr condition (SCAR 26-041) due August 22. Interim containment: 100 percent sort continues until scrap is under 0.5 percent for three consecutive lots. Current rate 1.5 percent.
- Replacement deburr fixture on order, quoted at 8 weeks. We told Meridian's SQE on the July 24 call we would pursue a 6 week expedite. No confirmation yet from the tool shop.
- Price adjustment on the console bezel family: we committed in the June QBR to hold current pricing through Q1 next year in exchange for the volume moving to 100 percent.

Closed this quarter:
- The label spec discrepancy from May, closed June 12.
- Returnable packaging count discrepancy, credited on the June invoice.

Notes: Meridian launch review for the crossover program is September 30. Their SQE, R. Alvarez, prefers written status updates two days before calls.

Give this an Anti-AI Polish Pass.
Make it sound like a sharp human operator wrote it, not AI.

Cut anything:
- repetitive
- overexplained
- generic
- padded
- performative
- vague
- too long
- trying too hard

Remove AI tells:
- repeated sentence patterns
- filler transitions
- corporate jargon
- vague claims
- motivational-poster energy
- anything that sounds written instead of said

Cut at least 20%.
Deliver only when it is clear, human, sharp, concise, and impossible to mistake for AI.