rescan-skipped

name	rescan-skipped
description	Re-score a run's skipped jobs and promote the eligible ones to `approved` for later applying. Recovers jobs wrongly dropped for location, a sparse JD, 1099, or seniority. Does not apply.
argument-hint	<run-id> [--jobs key1,key2,…]

Rescan Skipped — Recover Wrongly-Dropped Jobs

Re-score a run's skipped jobs and set eligible ones to approved. Never apply and never change the run's status — apply the promoted jobs afterward via the apply skill (apply run <run-id>) or the run page.

Setup

JOBPILOT_API=http://localhost:8000

Follow plugin/skills/shared/setup.md. Fetch the run: curl -fsS "$JOBPILOT_API/api/runs/<run-id>". Threshold = config.minScore (fallback data.autoApply.minMatchScore, else 70).

Select Targets

Targets are every status:"skipped" job; with --jobs key1,key2,…, restrict to those keys.

• Always leave (permanent) — only these: skipReason starting Already applied, CAPTCHA, or Payment required, or one stating a JD-cited citizenship/clearance requirement.
• Whole-run mode (no --jobs): also leave deliberate user choices — Removed by user, Not selected by user, User cancelled…, Max applications limit reached, Run paused by user.
• --jobs mode: reconsider every named target except the permanent ones.

Count the full target list up front and process every one. Below-threshold, zero-score, and no-skipReason jobs are all targets — the stored score came from the run that wrongly skipped them, so it's never a reason to skip the re-score. Don't cherry-pick the jobs already at/above threshold.

Per Job

1. Digest — parse the cached digest. Rich = non-empty techStack and requirements/responsibilities.
2. Re-read only when needed — if the digest is thin/empty, or the original skipReason was invalid (location/onsite, sparse JD, 1099, seniority), open the posting (browser_navigate + narrowed browser_snapshot; log in via plugin/skills/shared/auth.md if walled), rebuild the digest, and write it back so future rescans skip the browser:

curl -fsS -X PATCH "$JOBPILOT_API/api/runs/<run-id>/jobs/<key>" \
  -H 'content-type: application/json' \
  -d "$(jq -n --arg digest "$DIGEST" --arg desc "<posting text>" '{digest:$digest, description:$desc}')"

3. Re-score — every target gets a fresh POST /api/score-fit with {digest}; never reuse the stored matchScore. If confidence >= 0.7 and score is ≥10 from the threshold, trust it; else deliberate from strongMatches/partialMatches/gaps. A zero/low score with no skipReason (common at defense/federal employers) is not a disqualifier — only a JD-stated citizenship/clearance bar is (never infer from industry).
4. Decide:
   • Eligible and score >= threshold → promote (no apply):

curl -fsS -X PATCH "$JOBPILOT_API/api/runs/<run-id>/jobs/<key>" \
  -H 'content-type: application/json' \
  -d "$(jq -n --argjson score <0-100> --arg reason "<one line>" '{status:"approved", matchScore:$score, matchReason:$reason}')"

• Below threshold after a fair read → leave skipped, PATCH skipReason:"Below minimum match score (X < Y)".
• JD-stated citizenship/clearance found on re-read → leave skipped with that reason.

Eligibility

Same rules as auto-apply 2.2a. Seniority is never a skip — never drop a role for being below your level (Junior/Mid when your résumé is Senior) or for asking fewer years than you have; over-qualification is full marks on experience. Location/onsite, sparse JDs, and 1099/contractor work are never skips either — only a JD-stated citizenship/clearance requirement disqualifies.

Finish

Process every target before finishing — promoted + left-skipped + permanent must equal the target count. If any are unprocessed, keep going; don't report a partial pass as complete.

Print a short table (promoted vs left skipped, with reasons) plus the reconciliation (e.g. "228 skipped → 226 targets; 19 promoted, 207 left skipped"). Then point the user to apply run <run-id> (or the run page's Apply selected). Don't apply; don't change run status.