Evaluate Multiple Jobs in Parallel

Evaluates multiple jobs in parallel instead of one-by-one by spawning headless workers that each generate reports and tracker entries independently, then merges results. You can assess a week's worth of offers in one batch run instead of context-switching between them.

Run /career-ops batch with a list of URLs. 1. Conductor (headed mode) navigates portals in real time, extracting JDs. 2. For each job, a headless worker process receives the URL + JD text, evaluates it (blocks A-G), generates report .md + PDF + tracker line. 3. State is tracked in batch-state.tsv (pending, processing, completed, failed, rate_limited, paused_rate_limit). 4. Workers are independent: failure in job #5 doesn't affect job #6. 5. If interrupted, re-run resumes from batch-state.tsv (skips completed jobs). 6. After all jobs complete, merge tracker-additions/ into applications.md. Output: N report .mds, N PDFs, N tracker entries, and a summary of scores + statuses.

Conductor must run with --chrome flag to navigate portals. Each worker is a headless subprocess (command differs by CLI: claude -p, opencode run, codex exec, etc.; see AGENTS.md Headless / Batch Mode table). Batch state tracks processing metadata; lock file (batch-runner.pid) prevents double execution. Parallel mode uses --parallel N to control worker count (default: 3-5).