Quick Verdict
career-ops 1.22 is an MIT-licensed, local-first job-search operating system. It is not an unattended browser bot for mass applications. It places role discovery, structured evaluation, tailored resumes, open-question drafts, company research, contact preparation, and pipeline tracking in a local project, then lets coding agents such as Claude Code, Codex, and OpenCode operate on those files. The product boundary matters: it can scan, analyze, and draft, but the candidate reviews and submits every application.
It best fits technically confident candidates who will maintain a career profile, use a terminal, and inspect AI output. The current methodology is five scoring dimensions plus a holistic global score, not the old claim of six dimensions. The dimensions cover match, north-star alignment, compensation, cultural signals, and red flags, feeding a cited 1.0-to-5.0 judgment. career-ops is not marked as recommended because resumes, compensation, preferences, and application answers are sensitive, and local-first operation does not mean that every model inference stays on the device.
Best For
- Candidates comparing dozens or hundreds of roles with a consistent rubric and pipeline.
- Users comfortable with Node.js, Git, terminals, and a coding agent who can review generated files and diffs.
- People who want inspectable data files, methods, and automation rather than a closed job-search SaaS.
- Applicants tailoring resumes, questions, and interview stories while retaining final human approval.
- Not ideal for unattended auto-apply, users unable to verify facts, non-technical candidates, or environments requiring guaranteed offline inference.
Key Features
- Five dimensions plus holistic score: Evaluates match, direction, compensation, culture, and red flags before producing an evidence-led global score.
- Role scanning and deduplication: Discovers postings across Greenhouse, Ashby, Lever, and company career pages, then maintains pipeline integrity.
- Tailored resume and PDF drafts: Uses a real CV and job description to prepare an ATS-oriented candidate version for human fact and layout review.
- Question and email drafting: Drafts open-ended application answers, recruiter email, and contact messages without sending them.
- Drafts, never submits: Application mode returns editable answers to the user and never clicks the final submission or sends a message.
- Research and interview preparation: Organizes company evidence, role risks, compensation clues, and STAR stories for interviews and negotiation.
- Local pipeline: Stores profile, CV, configuration, reports, and tracking data in the user’s local project with a terminal dashboard.
- Multiple agent hosts: Uses open skill files with Claude Code, Codex, OpenCode, Qwen, Kimi, GitHub Copilot CLI, and related agents.
Use Cases
A candidate can record their verified experience, preferences, proof points, target roles, and exclusions locally, then scan for opportunities. Each role receives a five-dimension assessment and holistic score. Weak fits are filtered; stronger candidates proceed to deeper research, a resume draft, and application preparation. Before submission, the user checks the company, title, compensation, employment dates, skill level, claims, and every open answer so the model cannot improve apparent fit by inventing experience.
Batch processing should rank work, not make the final decision. Recruitment pages are untrusted input. They can be stale, incorrect, or contain prompt-injection text. An agent must not disclose a local CV, environment variable, or account credential because a page instructs it to do so. It should not install unknown dependencies, run arbitrary scripts, or inspect unrelated files. Contact and company research also produce leads rather than verified identity; confirm a person’s current role and public contact route through first-party or reputable professional sources.
Pricing
As of July 21, 2026, the current career-ops version is 1.22.0. The project is MIT licensed and states that career-ops itself is permanently free, with no paid tier, account, or telemetry. Users can initialize it with npx @santifer/career-ops init or clone the official repository. PDF generation can require Playwright and Chromium, while the dashboard and supporting components add local runtime and maintenance requirements.
“Free” applies to career-ops software. Total cost depends on the chosen agent, model or API, web research, browser automation, machine resources, and review time. Gemini CLI, Claude Code, Codex, and other hosts each have separate subscriptions, quotas, data terms, and rate limits. Back up profiles and pipelines before upgrades, read the changelog, and validate scripts and document formatting in a sample project.
Pros
- MIT licensing makes the rubric, scripts, and data structures inspectable and adaptable.
- Local-first files and no career-ops account or telemetry reduce dependence on a centralized applicant profile.
- Five dimensions plus a holistic judgment are easier to challenge than opaque keyword matching.
- The product explicitly preserves human editing and submission instead of defaulting to mass auto-apply.
- Covers discovery, evaluation, resumes, application questions, research, interviews, and tracking.
- Reuses existing coding agents and does not require one model vendor.
Cons
- Node.js, Git, terminal use, browser dependencies, and agent configuration create a meaningful setup burden.
- Local-first is not fully offline; the host may send prompts, resume excerpts, and page content to its model provider.
- Recruitment pages and third-party research can be stale, wrong, or adversarial.
- ATS resumes, compensation research, and scores cannot guarantee interviews or offers.
- An English-market method does not prove equal reliability for Chinese hiring sites, salary sources, or PDF conventions.
- Fast open-source iteration requires dependency security, upgrades, and custom configuration to be maintained.
Alternatives
| Tool | Best for | Main advantage | Watch for |
|---|---|---|---|
| Claude Code | Terminal users building their own career files and scripts | Strong general repository agent | A possible host, not a complete job-search methodology |
| Codex | OpenAI-centered file processing and automation | Integrated coding and task execution | Different account, model, and cloud-data boundary |
| OpenCode | Users seeking an open client and provider choice | Flexible host and model selection | Still needs career-ops or a custom job-search structure |
| Gemini CLI | Users already adopting Google’s developer tools | Open terminal client and Gemini integration | Cloud processing and quotas need separate review |
| GitHub Copilot | Candidates already in GitHub development workflows | Mature CLI and development ecosystem | Request billing and limits are outside career-ops |
FAQ
Does career-ops submit applications automatically?
No. It can scan, score, draft resumes, and prepare application answers, but it deliberately leaves editing and final submission to the candidate.
Does it currently use six scoring dimensions?
No. The current public method uses five scoring dimensions feeding a holistic global score. Report sections labeled A through F are organizational blocks, not six independent dimensions.
Does local-first mean no data ever leaves the computer?
No. career-ops keeps its own files locally and does not require its own cloud account, but the selected coding agent, model API, browser, and recruitment sites are independent processors.
Is career-ops completely free?
The MIT software has no official paid tier. Models, subscriptions, API calls, machine resources, network use, browser automation, and maintenance time may still cost money.
Can generated resumes and application answers be used unchanged?
They should not be. Verify dates, employers, projects, skills, compensation, outcomes, and tone. Remove every claim that cannot be supported and follow the employer’s application rules.
Does it work for Chinese job searches?
It can be adapted, but the current methodology and integrations are more aligned with English-language markets. Validate fonts, PDFs, portal extraction, compensation evidence, and rubric localization before expanding usage.
Bottom Line
career-ops 1.22 turns job search into an inspectable local operating process rather than a race to click more applications. Five dimensions, a holistic score, tailored material, research, and tracking share the same files. Its safety conditions are equally important: job pages are untrusted input, the model provider is a separate data boundary, AI must not embellish a resume, and the candidate performs the final submission. A technical user can pilot it on ten representative roles. Someone who does not want to maintain a terminal environment or only needs one resume edit will usually prefer a lighter tool.