Career Ops v1.32.0 is a local job-search workspace
Career Ops v1.32.0 turns a folder of instructions, scripts, personal files, and reports into an agent-guided job search. Paste a listing and it compares the role with your CV, writes an A-through-H evaluation, can make a tailored PDF, and records the application. Scanners cover common applicant-tracking systems and company pages. A Go terminal dashboard offers filters and sorting, while the web interface is labeled experimental alpha software.
Our measured checkout had 1,373 files, about 229,101 lines of source, and occupied 44.9 MB. Data is mostly Markdown, YAML, and TSV. JavaScript handles the workflows, Playwright handles browser and PDF work, and Go powers the terminal interface. That structure is inspectable and easy to back up, but users are taking on a small software system rather than adding one browser extension.
A 4.0 score cutoff makes it a filter, not an auto-applier
Career Ops recommends against applying below 4.0 out of 5 and says its score is a judgment across five dimensions rather than simple arithmetic. It researches fit, requirements, compensation, legitimacy, work authorization, contacts, and interview material. The agent ranks and drafts while the person decides. The README repeatedly says it never sends, submits, or performs the final click, including for application emails and form filling.
A successful install would not turn this into an unattended recruiting bot, and our install did not succeed after 30 seconds. Job seekers must review recommendations, tailored claims, portal actions, and every attachment. The disclaimer says AI can invent skills or experience. It also places responsibility for job-board terms on the user, which matters when Playwright checks postings or fills forms.
What happened when we ran it
Our sandbox cloned commit afc5275 and attempted npm installation with 3 CPUs, 8 GB of RAM, and no secrets. The install exited with code 1 after 30 seconds. Its final output said the process was running Node.js 18.20.8 and that Playwright requires Node.js 20 or higher. The postinstall command attempted to install Chromium with system dependencies, retried the same command, and ended with npm error code 1.
The log did not show a build or test result after that failure, so we do not claim either step passed. The repository scan found 20 CI workflow files, a Dockerfile, a Compose file, and a tests directory. Those signals do not override the failed setup on our box. commit afc5275 did not complete npm install in the stated container because its postinstall Playwright command rejected the Node.js version.
Package metadata declares Node.js 18 or newer, and the setup guide also lists Node.js 18+ as the baseline, with Node.js 20+ called out only for one CLI integration. Playwright 1.62.1 is pinned in the same package file. Our Node.js 18.20.8 result shows that following the written baseline can stop the general install before onboarding, even when the user does not yet intend to render a PDF.
Node.js 20 avoids the failure that Node.js 18.20.8 hit
Start with Node.js 20 or newer until the documented minimum and postinstall behavior agree. The quick path runs npx, clones a release, and installs dependencies. Manual setup adds a profile file, portals file, and CV, followed by the agent session. Chromium is described as a one-time PDF requirement, although postinstall invokes it during npm install. Go 1.21+ is optional unless the terminal dashboard matters.
Model access is a separate setup choice. Coding-agent login, provider API keys, OAuth tokens, or a local model endpoint may be involved, depending on the runner. The FAQ warns that an Anthropic API key can take precedence over an interactive subscription and produce usage charges. The 30-second lab failure happened before any of these credentials were needed.
Local files still send CV content to the chosen provider
The v1.32.0 privacy statement says the maintainers do not collect your files, while your CV and personal data are sent directly to the AI provider you select. Local storage helps with ownership and migration, but does not keep every detail on the machine. Users handling addresses, phone numbers, salary preferences, or work authorization should inspect the provider path before feeding the agent a full profile.
Open issue 3924 reports that the dashboard start command ignores an external data-directory setting and resolves reports against the repository root. The 44.9 MB checkout is easy to duplicate, while private data needs a deliberate backup. Browser-only Claude Code has another wrinkle: ignored CV, data, and report files do not persist in Git between fresh cloud sessions, so the setup guide tells users to move needed output to secure storage.
September 2026 activity is high, and issue 3887 is a release blocker
GitHub showed 70,308 stars and 393 combined open issues and pull requests. The repository was pushed on September 6, 2026, while Career Ops v1.32.0 and web v0.10.0 were released on September 3. Earlier paired releases appeared on August 31. That pace and the same-day issue discussion indicate active maintenance; the combined open count should not be mistaken for 393 confirmed defects.
Open P1 issue 3887 is serious enough to change our recommendation. It reproduces a v1.32.0 path where generating a cover letter replaces the tailored-CV row for the same report. The apply flow can then select the cover letter as the CV. Until that is fixed in a release, check the actual filename and document before any upload. A 30-second install failure is annoying; sending the wrong document to an employer deserves the harder stop.
The 1,373-file system rewards careful, technical users
Career Ops is an editable operating manual backed by code, rather than a finished consumer service. Its modes, scanners, local files, and agent instructions give a technical candidate direct control over how a search is evaluated. The trade is ownership: runtime versions, model credentials, personal-data handling, portal rules, and generated files all need attention. Start with Node.js 20+, keep the human approval boundary, and verify every attachment. A hosted tracker is the better choice when that process sounds excessive.