进阶 career-ops.org 2026-10-16 02:37:12 · 44 阅读

第24章 Career-Ops Pipeline 模式:自动化批量处理求职申请

ModespipelineReads URLs queued in data/pipeline.md and runs the full auto-pipeline against each one. Use it when you collected ten or twenty listings during the day and want to crunch them in a single session.Updated October 14, 2026What it does

pipeline mode iterates over the URLs in data/pipeline.md (one per line) and invokes auto-pipeline against each. Reports, PDFs, and tracker entries are written for every listing that produces a valid evaluation. Failed scrapes are logged and skipped.

When to use it

Use pipeline when you have a backlog of URLs to process. Drop them in data/pipeline.md, invoke /career-ops pipeline, and step away.

What a run looks like

You drop URLs into data/pipeline.md and the mode processes the whole inbox in one pass. First it runs a zero-token liveness sweep to drop expired postings, then it evaluates each surviving URL through the full auto-pipeline and ends with a summary table of what it found.

Pipeline run$ /career-ops pipeline

Liveness sweep: 8 pending URLs 6 live · 2 expired (moved to Processed, marked Discarded)

Processing 6 live URLs (one at a time)...

| # | Company | Role | Score | PDF | Recommended action | |---|---------|--------------|-------|-----|--------------------| | 1 | Acme | Staff AI Eng | 4.3/5 | done | apply | | 2 | Globex | ML Platform | 3.1/5 | skip | review | | 3 | Initech | AI PM | 2.4/5 | skip | skip |

Each processed URL moves from the ## Pending to the ## Processed section of the file, so the inbox stays a live record of what is left.

When to reach for it

Reach for pipeline when you have collected a backlog of listings during the day and want to crunch them in a single session. How it handles the queue:

The ## Pending section accepts any shape — a bare pasted URL works, or you can add company, role, location, and comp columns, and the mode reads them all. The liveness sweep runs before the per-URL loop, so a stale inbox of eight URLs does not cost eight wasted evaluations. LinkedIn URLs skip the liveness sweep and get one browser attempt in the per-URL loop instead. If that attempt leaves the description unavailable, for example a page stuck on a loading skeleton, the entry stays in Pending marked [!] and you are asked to paste the job text or give an employer or ATS link for the same role. A failed LinkedIn load is never treated as a closed job. The extraction tool decides the concurrency. With the default browser extractor (Playwright MCP) it processes URLs one at a time, because workers sharing one browser session can cross-contaminate and evaluate the wrong posting. Only when every worker uses the isolated CLI extractor (scan.extractor: cli) or a non-browser fallback, each with its own process and session state, can three or more URLs run in parallel, one worker per URL. PDFs are generated only when a score clears auto_pdf_score_threshold (default 3.0); raise it to skip PDFs for marginal roles and generate them on demand later with /career-ops pdf.

Related

auto-pipeline auto-pipelineDefault flow when you paste a job description URL or raw text. Runs the A–H evaluation, generates the report, exports a tailored PDF, and writes a tracker entry — without manual intervention between steps.applyLive application assistant. Reads the open-ended questions Greenhouse, Ashby and Lever forms ask, and drafts answers grounded in your CV and the JD.

评论 (0)