Your AI career command center

After a layoff you end up re-explaining your whole situation to a blank AI chat every single time you want help with a resume bullet, an outreach note, or an interview answer. Each new chat forgets who you are, so the first ten minutes go to context and the output still comes back generic. A command center fixes this permanently. You load your situation once into a persistent AI Project, and from then on every chat, for every part of the search, already knows your history, your targets, your wins, and how you write. Among people who already use AI at work, 90 percent say it saves them time and 85 percent say it helps them focus on their most important work (Microsoft and LinkedIn, 2024 Work Trend Index; self-reported by AI users, not measured output). This course is not another prompt list. It is the connective tissue underneath the prompts: one truthful Context Pack, one guardrail block, and one master tracker that every module reads from and writes to, so the modules compound into a single self-diagnosing pipeline instead of scattered chats that each start cold. That is the whole difference from a tool-by-tool tactics course, and it is why this one setup is worth doing carefully.

Set it up in whichever tool you already pay for; the moves are identical. In Claude, click New Project, add your files under Project knowledge, and paste your operating rules into the project's Custom instructions. In ChatGPT, create a Project, drop the same files in, and set the project instructions. Both attach those files and rules to every chat inside the project, which is the entire point: you stop re-explaining yourself. On a free plan with no Projects you are not blocked. Use the free-tier mega-prompt in the template, which packs the whole Context Pack and the guardrails into one message you paste at the top of each new chat, so a plain chat behaves like a loaded project for that session. Into the knowledge, or the mega-prompt, go the five files of your Career Context Pack. career-facts holds your verifiable roles, dates, scope, and metrics. target names the two or three roles you are aiming at, with titles, comp range, and dealbreakers. wins holds eight to twelve accomplishment stories in situation-task-action-result form, every number one you can defend. voice holds two or three real messages you have actually sent, so drafts sound like you. situation holds the honest, non-defensive version of what happened plus your own runway and timeline, which stays private and is only there to help the system plan around your reality.

The instructions matter as much as the files. In the project's custom instructions, define the assistant as your career operations partner and set hard guardrails: use only facts from the files or what you paste, never invent an employer, title, metric, or date, mark anything missing as needs verification, match your voice on anything you will send, stay calm and direct with no false reassurance and no guarantees about jobs or timelines, and treat any money or legal point as general guidance to verify against the real documents. These rules are what keep the system honest. A tool that invents a flattering metric will get you caught in an interview or a reference check, so you build the truthfulness in at the root.

The second half of the command center is one master tracker: a single spreadsheet or database, Google Sheets, Airtable, or a Notion database all work, that becomes the memory of the search. One row per opportunity, with columns for company, role, source, date found, fit score, stage, next action, next-action date, and contact. Every later module reads from or writes to this one table, which is exactly what turns eight separate tricks into one operating system. Build the shell now; the modules ahead fill it.

• Create one Project in Claude or ChatGPT named for your search, and confirm your plan lets you add files and custom instructions.
• Build the five Career Context Pack files from the template and upload them to the project's knowledge.
• Paste the operating instructions into the project's custom instructions and adjust them to your situation.
• Test it: in a fresh chat ask, based on my files, what are the three role types I should focus on and why, then fix any file the answer reveals is thin.
• Create the master tracker with the nine columns, and paste its link into the project instructions so the AI knows where your pipeline lives.

Same question, two different worlds. In a blank chat, "help me write a resume bullet about improving a process" returns "Leveraged cross-functional synergies to streamline operational workflows," a line that belongs to anyone and survives no follow-up question. Inside a loaded project, the same question pulls the real story straight from wins.md, that vendor onboarding went from six weeks to nine days after the intake checklist was rebuilt, and drafts it in the plain, direct style of the two emails in voice.md. Ten seconds instead of ten minutes of re-explaining, and a bullet specific enough to defend out loud under a follow-up. The difference is not a cleverer prompt, it is the context the project already holds.

• Writing all five context files from memory in one sitting, which produces vague claims and rounded-up numbers you cannot defend. Pull dates, scope, and metrics from your actual resume, old performance reviews, and sent emails. If you cannot verify a number, leave it out or mark it needs verification.
• Uploading the files but skipping the custom instructions, so the AI still invents flattering details when your files run thin. Paste the full guardrails block into custom instructions. The "never invent an employer, title, metric, or date" rule is what makes every later module trustworthy.
• Letting the tracker live in three places, a doc, a notes app, and a half-built spreadsheet, so nothing is the source of truth. One table, nine columns, one link pasted into the project instructions. Delete or archive every duplicate today.
• Assuming you need a paid plan and stalling because Projects are locked on your tier. Use the free-tier mega-prompt from the template. It carries the full Context Pack and guardrails in one pasted message, so a plain chat runs as a loaded command center for that session, no upgrade required.