Career-Ops turns an AI coding CLI into a deliberate job-search pipeline

A job search needs triage, not just automation

A job search quickly produces a familiar operational problem: job descriptions arrive from several portals, decisions are scattered across notes or spreadsheets, and every promising role seems to demand a fresh CV. The README for Career-Ops presents the project as a local command center for that workflow. Rather than treating automation as a way to apply everywhere, it is designed to help a candidate narrow a large pool to roles worth pursuing.

That distinction shows up in the documented recommendation threshold and in the project’s explicit human review boundary. The system can evaluate, generate materials, and prepare application-related work, but it is not intended to submit applications automatically. For builders experimenting with agent-assisted personal workflows, this is a useful framing: use an agent to assemble evidence and reduce repetitive work, while keeping consequential decisions with the person using it.

Career-Ops is aimed at people comfortable working in a repository and an AI coding command-line interface. The README documents support for Claude Code, OpenCode, and Gemini CLI, with a standalone Gemini API evaluation script also available. Its inputs are deliberately personal: a Markdown CV, a YAML profile, and optional proof points. The stated goal is for the evaluation logic to use that context when comparing a candidate with a job description.

From a pasted role to a tracked decision

The central interaction is intentionally simple. A user can paste a job URL or job description into the CLI, then Career-Ops detects the input and runs its documented auto-pipeline. That pipeline classifies the role into an archetype, evaluates the fit against cv.md, and produces three artifacts: a Markdown report, a PDF CV, and a tracker entry.

The evaluation is described as an A-F system with weighted dimensions. Its six documented blocks cover a role summary, CV match, level strategy, compensation research, personalization, and interview preparation. The project also describes an interview story bank based on STAR plus reflection. STAR is a common interview structure: situation, task, action, and result. Here, the repository proposes accumulating reusable stories across role evaluations instead of writing behavioral examples from scratch each time.

The workflow extends beyond a single listing. A scanner can search configured company portals and job boards, while batch processing uses parallel claude -p workers for multiple offers. The README says the portal configuration includes more than 45 companies and 19 queries, and supports sources such as Ashby, Greenhouse, Lever, Wellfound, Workable, and RemoteFront. Those defaults are customizable through a YAML template, which matters because a targeted job search usually has a different company list and vocabulary for each candidate.

Tracking is treated as a data-maintenance problem rather than a side effect. The manifest confirms scripts for status normalization, deduplication, merging tracker data, validation, and a CV sync check. The README describes Markdown tables, YAML configuration, and TSV batch files as the project’s data layer. This file-based approach makes the system inspectable and editable with ordinary tools, but it also asks users to be comfortable maintaining files in a local project directory.

A repository of modes, templates, and generated outputs

The documented structure separates instructions from user data. The modes/ directory holds 14 skill modes and shared context, covering activities such as single-offer evaluation, PDF generation, scanning, and batch work. Templates provide a CV HTML template, portal configuration example, and canonical application states. Generated reports, PDFs, and tracking data are placed in directories marked as ignored by Git.

That separation is one of the practical ideas in the project. A reusable workflow needs stable scaffolding, while a job search contains highly personal and changing information. Keeping profile data, CV content, reports, and outputs distinct from the modes and templates makes customization more manageable. The README goes further by encouraging users to ask the AI CLI to adapt archetypes, scoring weights, scripts, and modes. That is powerful, but it creates a tradeoff: users should review changes to the workflow logic just as carefully as they review generated application materials.

Career-Ops also includes a terminal dashboard written in Go. The documented build command produces a career-dashboard binary that reads the parent project directory. Its features include filter tabs, sorting modes, grouped or flat views, previews, and inline status changes. For a workflow that already stores its state in local files, a terminal UI is a fitting layer: it provides a focused view without requiring a hosted database or web application.

Confirmed pieces of the stack

The supplied package.json confirms a Node.js project with scripts for setup checks, scanning, PDF generation, tracker maintenance, updating, and Gemini evaluation. It lists Playwright, js-yaml, dotenv, and Google’s generative AI SDK as dependencies.

Playwright has two documented roles: browser automation for scanning career pages and browser-based PDF generation from the HTML CV template. js-yaml supports the profile and portal configuration format. The Google SDK underpins the optional Gemini API path. The README also identifies Go, Bubble Tea, and Lipgloss as the dashboard stack, with the latter two providing terminal interface components and styling.

There is a sensible privacy and safety tradeoff in the README’s local-first positioning. The project says personal data stays on the user’s machine and is sent directly to the chosen AI provider, rather than to a hosted Career-Ops service. At the same time, that means users need to understand which provider they configure and verify every generated claim before submitting it. The README explicitly warns that model output may be inaccurate and that users must comply with the terms of the job portals they use.

For builders looking at agentic workflows beyond software development, Career-Ops is a concrete example of turning a coding CLI into a structured, file-backed operating system for a personal process. Inspect the repository at https://github.com/MuLIAICHI/career-ops-ui.