resume

name	resume
description	Resume an interrupted or paused JobPilot run by id. Re-flips the run to in_progress and replays the apply loop on any remaining approved jobs without re-asking for fit confirmation.
argument-hint	<run-id>

Resume — Continue an Interrupted Run

Resumes a paused or interrupted Run by replaying the apply loop on jobs that are still approved (or pending if approval was implicit). The user already approved the fit when the run was first launched, so no re-confirmation gate.

Live view: http://localhost:8000/runs/<run-id>.

Setup

Follow plugin/skills/shared/setup.md to load profile, resume, credentials. Check the web app is up:

JOBPILOT_API=http://localhost:8000
curl -fsS "$JOBPILOT_API/api/health" >/dev/null || { echo "JobPilot web is down. Start it with 'bun run dev'."; exit 1; }

Phase 0: Resolve Run

Argument is <run-id>. If missing, list candidates and ask:

curl -fsS "$JOBPILOT_API/api/runs" \
  | jq -r '.data[] | select(.status=="paused" or .status=="interrupted")
           | "\(.runId)\t\(.status)\t\(.source)\t\(.query)"'

Fetch the run + jobs:

RUN_ID="<run-id>"
RUN=$(curl -fsS "$JOBPILOT_API/api/runs/$RUN_ID")

Verify status is paused or interrupted. If completed or failed, stop: "Run is already . Nothing to resume." If in_progress, stop: "Run is still in progress. If the agent crashed, wait for the auto-reconciler (5 min) or stop the run from the UI first."

Refuse to resume if there are no jobs with status === "approved":

APPROVED=$(echo "$RUN" | jq '[.data.jobs[] | select(.status=="approved")] | length')
[ "$APPROVED" = "0" ] && { echo "No approved jobs left to apply. Mark the run completed from the UI."; exit 0; }

Phase 1: Re-open the Run

PATCH status back to in_progress:

curl -fsS -X PATCH "$JOBPILOT_API/api/runs/$RUN_ID" \
  -H 'content-type: application/json' \
  -d '{"status":"in_progress"}'

Read config.maxApplications from the run for the stop condition below:

MAX_APPS=$(echo "$RUN" | jq -r '.data.config.maxApplications // empty')

Phase 2: Replay Apply Loop

For each job where status === "approved", score-descending, run the exact same per-job flow as the apply skill's Phase 4 (see plugin/skills/apply/SKILL.md sections 4.1 through 4.9):

1. 4.1 Mark Applying — PATCH job to applying.
2. 4.2 Navigate + Find Apply — browser_navigate, then browser_snapshot the header to find the Apply control and click its ref; browser_snapshot the form to enumerate fields and refs.
3. 4.3 Tailor Resume — read the cached digest from the Job row, invoke the tailor-resume skill with it. On failure POST /result outcome:"failed", failReason:"No tailorable resume base".
4. 4.4 Fill Forms — follow plugin/skills/shared/form-filling.md; upload the tailored variant.
5. 4.5 Pre-Submit Review — skip unless config.maxApplications === 1.
6. 4.6 Submit — submit; browser_wait_for, then a narrowed browser_snapshot for the success or error result.
7. 4.7 Record Result — POST /api/runs/$RUN_ID/jobs/<key>/result with outcome:"applied"/"failed"/"skipped". Atomic: updates the Job, creates Application, marks queue consumed, recomputes summary.
8. 4.9 Limit — if MAX_APPS set and summary.applied >= MAX_APPS, POST /result outcome:"skipped", skipReason:"Max applications limit reached" for each remaining approved job and end the loop.

The /result endpoint preserves the run's original source ("apply" vs "auto-apply") on the created Application row automatically — no separate source-passthrough needed.

Between jobs: honor user Stop

Re-fetch the run between jobs and exit cleanly if the user stopped it:

STATUS=$(curl -fsS "$JOBPILOT_API/api/runs/$RUN_ID" | jq -r '.data.status')
if [ "$STATUS" = "paused" ]; then
  # POST /result outcome:"skipped" skipReason:"Run paused by user" for each remaining approved job, then stop
  exit 0
fi

Phase 3: Summary

NOW=$(date -u +%Y-%m-%dT%H:%M:%SZ)
curl -fsS -X PATCH "$JOBPILOT_API/api/runs/$RUN_ID" \
  -H 'content-type: application/json' \
  -d "$(jq -n --arg t "$NOW" '{status:"completed", completedAt:$t}')"

Print a summary table and the run link http://localhost:8000/runs/<RUN_ID>. Suggest re-running the auto-apply skill in retry-failed <RUN_ID> mode if any jobs failed.

Rules

1. No new confirmation gate. The user already approved the fit when the run was first launched.
2. Preserve source when recording applications — a resumed apply run still records source:"apply", not "resume".
3. Idempotent. Resuming the same run a second time should be a no-op when no approved jobs remain.
4. The Run is the audit trail. PATCH after every state change so the SSE viewer reflects reality.