suxrobgm/jobpilot/search

Job Search

Search a single board (picked by the user when launching the campaign) and rank results by qualification fit against the resume.

Setup

1. Follow ../_shared/setup.md.
2. Parse and strip the flags; the rest is the free-text query.

• --board <domain> - required (e.g. --board linkedin.com).
• --max-jobs <N> - optional cap on results to rank. Absent = unlimited (rank until the board is

exhausted).

• --campaign <campaign-id> - campaign to save results to (Phase 5). The web UI passes it; if

absent, match the latest source:"search", status:"in_progress" campaign on the query, else create one (a source:"search" create requires config.resumeId - default to the profile's primaryResumeId).

1. Resolve the board:

``bash jobpilot-api GET /api/job-boards ``

Pick the entry whose domain is <domain>. If none matches, abort with: "Board <domain> is not configured. Add it on /boards or run again with a different --board." When a --campaign id was given, first command it to failed with POST /api/campaigns/<id>/status {"status":"failed"}.

Phase 1: Parse Query

Extract title/role, keywords, location, other preferences (e.g. "no startups", "FAANG only", salary). If vague, ask before searching.

Phase 2: Search the Board

1. browser_navigate to the resolved board's searchUrl.
2. Follow ../_shared/auth.md to log in proactively.
3. Fill the search fields and submit.
4. Take a browser_snapshot narrowed to the results list (per ../_shared/browser-tips.md) and

read { title, company, location, url, postedAt } per row. While under --max-jobs (or always, when it's absent/unlimited), scroll/paginate per Pagination & infinite scroll in ../_shared/browser-tips.md until the cap is met or results run dry.

1. Take the first --max-jobs results (or all of them when unlimited); if fewer after paginating,

take what's there. Per row:

• Listing preview suffices (the normal case) → rank from the row; no per-job navigation.
• A brief description is needed for the ranked table and the preview lacks one → delegate

that row to the job-scorer subagent with mode:"score" and minMatchScore:0 (so nothing is auto-skipped - search keeps every result for review). It opens the posting, scores, and saves the Job row in isolated context. Such rows are already saved - exclude them from the Phase 5 bulk save.

Phase 3: Exclude Previously Applied

Run the applied-check (../_shared/campaign-flow.md) per result. If .applied, tag with "Previously Applied" (note .match.kind). In Phase 5, create these rows as pending, then record skipped through /jobs/<key>/result; do not offer them for apply.

Phase 4: Fit Review

Score each non-applied result server-side, like every other campaign skill. Build the brief (../_shared/job-brief.md) and write {"brief": <brief>, "minScore": 0, "resumeId": "<config.resumeId>"} to $JOBPILOT_TEMP/fit.json (drop resumeId when the campaign has none; the server uses the primary):

jobpilot-api POST /api/score-fit --data @"$JOBPILOT_TEMP/fit.json"

Use .score when .verdict is trust; on deliberate, adjust it from strongMatches/partialMatches/gaps. minScore: 0 because search keeps every result for the user to review.

Phase 5: Save Results to the Campaign

Save every result as a Job on <campaign-id> so it appears on the campaigns detail page. Don't offer apply/search-again commands - the user applies from there. Use a stable, shell-safe key per result (slug of company-title + rank, no spaces).

Carry the brief you scored from (../_shared/job-brief.md). Per result, write {key, title, company, location, url, board:"<domain>", matchScore, matchReason:"<one-line verdict>", status:"pending", brief} (brief as a JSON string) to $JOBPILOT_TEMP/job-<key>.json, then:

jobpilot-api POST /api/campaigns/<campaign-id>/jobs --data @"$JOBPILOT_TEMP/job-<key>.json"

Previously-applied results (Phase 3) → create as pending, then POST /jobs/<key>/result with {outcome:"skipped",skipReason:"Already applied (<kind>)"}. Then close the campaign:

jobpilot-api POST /api/campaigns/<campaign-id>/status --data '{"status":"completed"}'

Phase 6: Hand Off

Print a compact ranked table, then link to the campaign - nothing else:

## Saved <N> jobs · "[query]" - review and apply at $JOBPILOT_WEB/campaigns/<campaign-id>
 
| # | Score | Title | Company | Location |
|---|-------|-------|---------|----------|

Rules

The shared campaign rules (../_shared/campaign-flow.md) apply throughout. On top of them:

1. Exactly one board per campaign - the --board flag is required and the skill targets only

that board.

1. Handle rate limiting - if blocked, note it and continue.
2. Deduplicate within the board.

Read ../_shared/browser-tips.md for large pages, popups, and browser best practices.