Job Deck

A local agentic toolkit for automating and continuously calibrating your job search.

Job Deck helps with resume improvement, opportunity research, evidence-based evaluation, feedback, and progress tracking. It does not submit applications or make personal, legal, or employment decisions for the user.

Fictional demo data. No candidate information is shown.

Paste this into a new Codex task:

https://github.com/prof2k/JobDeck — I want to use this as my job-search agent.

That is enough intent for Codex to clone the repository, enter it, load its instructions, name the task Job search agent, and continue by asking for the current resume when one was not already attached. The user should not have to request setup, repeat the job-search request, or operate the repository.

The same link can start the workflow in ordinary ChatGPT. When a host cannot run the local dashboard but can create files, Job Deck delivers each completed batch as a standalone HTML report instead. The user does not need to install a runtime or handle a data file.

Give Codex this repository and say what you need. It will clone the toolkit when needed, read AGENTS.md, and continue with your request. If the folder is already open, just talk naturally.

• Attach your current resume and answer the few questions Job Deck cannot infer.
• Confirm the job directions it recommends; the first search then starts automatically.
• Ask naturally whenever you want to improve your resume, revisit a report, or change the search.

There is no setup command or fixed onboarding session. Job Deck checks what is already available, creates its private local workspace quietly, and keeps moving through resume intake, calibration, search, and reporting. You never need to name the next stage or repeat a request.

Each fresh Job Deck workspace starts with a blank candidate profile. It does not reuse assistant memory, earlier chats, saved job-search settings, sibling projects, or other personal files unless you explicitly identify the source you want imported. After you provide information here, this workspace can remember its own confirmed state for later Job Deck sessions.

• Checks an existing resume for material red flags without forcing a rewrite.
• Improves or helps create a resume when explicitly requested, without inventing facts.
• Reconciles the latest candidate evidence, preferences, feedback, and search history before each new search.
• Learns role, level, geography, work arrangement, relocation, compensation, authorization, and other preference boundaries without treating compatible location choices as mutually exclusive.
• Produces exactly 15 verified calibration roles before moving to exact batches of 50 additional roles by default; any number the user requests overrides 50.
• Searches the web through dynamically parallel, non-overlapping research lanes while reporting only roles that survive the user's hard gates and remain live on an official application page.
• Verifies official application pages and prevents duplicate recommendations.
• Shows requirement-level evidence, risks, gaps, and application links.
• Records direct feedback and explicit application progress for later searches.

Applications remain manual. Opening an official application page means Opened; a job becomes Applied only after an explicit user action.

1. Share a resume or ask Job Deck to help create one.
2. Answer only the missing search decisions in a short guided batch.
3. Confirm the titles, level, and boundaries Job Deck recommends.
4. Job research and report preparation begin automatically and in parallel.
5. Review exactly 15 calibration roles in the local dashboard or standalone HTML report.
6. Tell Job Deck which roles stand out. The next search automatically produces exactly 50 additional verified roles unless you request another number.

All candidate information, resumes, reports, progress, research history, and logs belong under private/. Only private/README.md is safe to track; every other path inside that directory is excluded from Git.

The tracked repository contains reusable instructions, living standards, fictional examples, privacy guards, and source code. The agent creates missing private files directly from the contract in AGENTS.md as they become useful.

• AGENTS.md — the central operating, search, privacy, reporting, logging, and maintenance contract.
• standards/ — living evaluation, resume, and report requirements.
• examples/ — fictional, non-candidate data for demonstrations and testing.
• private/ — ignored candidate data and local runtime state, plus one safe tracked folder guide.
• private/agent-workspace/ — compact, regenerable agent context such as the pre-search brief.
• src/server/ — Go server, SQLite schema, API, and tests.
• src/web/ — Vue interface source, design contract, and committed production output.
• src/static-report/ — self-contained single-report fallback used when the local dashboard cannot run.

The root jobdeck executable is created locally when a compatible runtime is already available. It is a derived artifact and is not distributed with the toolkit. Report delivery never requires the user to install Go, Bun, or Node.js.

Inside the private workspace, user-profile.md records factual candidate evidence, search-preferences.md records the current job-search contract, and user-preferences.yaml records how the user wants Job Deck itself to operate.

Job Deck binds to 127.0.0.1, keeps operational data under private/, and does not host candidate information. External research still uses the active agent's configured web tools and privacy settings.

The toolkit must not submit applications, complete assessments, contact recruiters, or answer personal or legal questions on the user's behalf.

Development details and the selective-build matrix live in src/README.md. Normal resume, preference, search, report, and logging work requires no source build.

Job Deck is available under the MIT License.