Job Triage.

The problem

Reading a job posting and deciding whether you actually fit takes real attention, and most of the postings you open are not worth that attention. The honest answer to “should I apply” is usually no, but it is buried under boilerplate, padded requirements, and the occasional scam.

What I built

Job Triage reads the job posting you’re already viewing, scores it against your résumé, and returns APPLY, MAYBE, or SKIP with reasons. On qualifying roles it drafts a cover letter, talking points, or a recruiter message grounded in your real experience. It runs free and offline out of the box; adding your own API key turns on résumé-aware AI verdicts and drafting. Built as a Manifest V3 extension with no framework and no build step.

Engineering highlights

• One pluggable judge behind a single contract, so the tier is a setting, not a rewrite. The offline keyword tier is also the fallback, so the tool never leaves the user stuck.
• A deterministic scam gate the model cannot override: pay-to-apply and fee posts are forced to SKIP by code, on top of any judge's output.
• Rubric-as-arithmetic prompting that lets a cheap model judge like an expensive one: stepwise method, defined terms, named failure modes, one worked example.
• Prompt caching for cost: the résumé sits in a stable prefix and only the job text changes per call, so repeat judgments bill a fraction of the input.
• API key isolated in the background service worker; it never touches the page.
• Editable drafting voice with separate, non-editable honesty rules, so a user can retune the tone but can't make it claim a skill the résumé doesn't show.