Your First Job

Once the CareerRat terminal workspace is open (run careerrat start; it needs Python 3.11 or newer) and your agent is live in the chat, use it the way a real candidate would. Here is the fastest path to confirm the whole stack, including the CLI, skills, agent reasoning, and your saved jobs, is working end to end. On a brand new install the chat starts with setup on its own, so step 1 may already be underway.

The 5-step test loop

1. Let it onboard you

The agent builds your profile by asking a few questions. To test without using real details:

set me up with a quick sample profile so I can test the flow

2. Paste a job posting and ask it to evaluate

Copy a job description from anywhere (a posting URL, raw JD text, or the bundled sample at examples/sample-jobs/) and paste it into the agent chat. Say:

evaluate this

You should get a structured verdict:

GATE: KEEP — strong overlap with role targets
FIT: high 88 — ...
COMP: clear — posted range covers target_base
ACTION: apply-now

Watch for activity lines while it works, reading the posting and maybe searching, each settling before the verdict prints.

This is a real read of the posting body against your config — not a keyword match. If the posting is a poor fit you will get GATE: CUT with a reason.

Here is the workspace with a job picked and the agent's answer beside it. The Fit report tab holds the same requirement-by-requirement read. See The terminal workspace for the tour.

3. Ask it to tailor

write a resume and cover letter for this

The agent builds honest artifacts from your evidence bank. It will not invent facts. If you used the sample profile from step 1, the output will be illustrative rather than personalized.

4. Paste a recruiter email

Copy any recruiter message (even a generic one) and paste it. Say:

draft a reply

The agent routes to the email-comms skill, drafts a reply in your writing style, and tracks the thread.

5. Check your saved jobs

Look at the jobs list on the left of the workspace and watch the application appear and change status. Click a job to see its details. Press F6 to move between the jobs and the chat. Everything the agent does writes through the same canonical data layer as the CLI, so the list stays in step with the chat. The assistant only speaks to ask a question or hand you a result.

If all five land, CareerRat is working end to end.

Test search discovery too

Ask CareerRat to search for a real role family and location. Built-in public job-board sources run alongside broad AI open-web discovery, and CareerRat finds additional specialist boards and employer pages from your saved roles. A credible AI result with no readable full description should still appear as AI · unverified, with its title, company, link, and visible search evidence. Choose Evaluate to verify liveness and capture the full posting before CareerRat treats it as application-ready.

The terminal search worker skips browser-backed saved searches and does not open a login window for them. Built-in public sources and AI open-web discovery can still run. Ask the agent to research a login-gated site separately when you need results from it.

While the search runs, move between Inspector tabs or keep talking to the agent. Activity shows its progress in the background. If the workspace is interrupted, reopen it to see the saved status and retry rather than a false completion.

Using the bundled sample

The repo ships a fictional sample candidate ("Riley Chen") and sample job in examples/:

examples/
  sample-candidate/   ← illustrative candidate config files
  sample-jobs/        ← a sample JD to evaluate
  demo-workspace/     ← a pre-seeded tracker for demos

You can point the agent at the sample JD directly:

evaluate the job in examples/sample-jobs/

Let the agent run the test itself

If you prefer to hand it off entirely, open your agent in the repo root and say:

read the README and AGENTS.md, set yourself up, and walk me through testing CareerRat end to end

The agent will self-direct through the full loop.

Expected pre-setup behavior

Until ingest-profile has run, careerrat doctor reports that local candidate setup is incomplete and lists the .careerrat/candidate/*.yml files to create. That is the expected pre-setup state, the agent fills them in during onboarding. It is not an error.

doctor also flags any candidate file that still carries unedited template content (the "Jane Candidate" example persona, placeholder tool names, and the like) so you can catch a half-personalized profile before it skews your job evaluations.