Job Triage

Thin wrapper around scripts/outlook-triage.js that runs a search, classifies results locally, and presents them as a table. The underlying script requires a Chrome instance running with --remote-debugging-port=9224 and Outlook Web already logged in.

Repo location

Same resolution as /job-apply: $AI_JOB_AGENT_ROOT → ~/.claude/skills/ai-job-agent/ → REPO_PATH marker file → ~/ai-job-agent/.

Prerequisites

Before doing anything, verify Chrome is reachable:

curl -s http://127.0.0.1:${OUTLOOK_PORT:-9224}/json/version >/dev/null && echo OK || echo MISSING

If MISSING, tell the user:

Start Chrome with remote debugging and open Outlook Web first:

/Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome \
  --remote-debugging-port=9224 \
  --user-data-dir=/tmp/outlook-triage-profile

Then navigate to outlook.office.com and log in. Rerun /job-triage.

Stop until they confirm.

Workflow

1. Run the search

cd "$AI_JOB_AGENT_ROOT"
node scripts/outlook-triage.js --port "${OUTLOOK_PORT:-9224}" search "$QUERY"

Default query: application status. If the user passes a different query via $ARGUMENTS, use that instead.

The script returns a JSON array of result objects with idx, aria, text, unread, flagged, pinned.

2. Classify each result

Run each text through these regex rules in order — first match wins:

Safety: if the sender contains @missouri.edu or @umsystem.edu (and not mailer/noreply/bot), force class = other — this matches the Mizzou-human safeguard already in the Outlook pipeline.

3. Render counts table

4. Render detail table (first 20)

Keep each row ≤ 120 chars.

4.5. Cross-reference with outreach-log.csv (auto-detect replies)

After classifying the inbox, read $AI_JOB_AGENT_ROOT/outreach-log.csv and try to match each unread email to a prior cold-email row by sender email (the to_email field in outreach-log is the recipient of the original send — when they reply, they become the sender of the inbound email).

Extract the sender email from each result's aria label or extracted body (use outlook-triage.js extract <idx> if needed for the headers). Match against outreach-log.csv.to_email (case-insensitive).

For every match where the row's replied_at is currently empty:

Render the proposed updates as a confirmation table:

🔁 Reply auto-detection — 3 matches in outreach-log.csv

| # | Recipient            | Original sent | Inbound class      | Proposed status     |
|---|----------------------|---------------|--------------------|---------------------|
| 1 | paul@gft.com         | 2026-04-10    | interview_invite   | interview_scheduled |
| 2 | zv@worley.com        | 2026-04-15    | rejection          | replied             |
| 3 | h.mueller@siemens... | 2026-04-22    | recruiter_outreach | replied             |

Apply all? Apply some (numbers comma-separated)? Skip?

On confirm, edit outreach-log.csv directly via the Edit tool — find each row by to_email, set replied_at to the current ISO timestamp, set status to the proposed class, and append a one-line note to notes. This skips /job-status because outreach-log is a different file from application-tracker.csv (and /job-status operates on the latter).

Print a summary:

Skip this step entirely if outreach-log.csv doesn't exist (cold-email setup wasn't completed).

5. Offer next actions (don't do them automatically)

Ask the user:

• "Extract one to read in full? (pass an index)"
• "Mark the confirmations as read? (all / one / none)"

If they pick extract:

node scripts/outlook-triage.js --port "${OUTLOOK_PORT:-9224}" extract <idx>

If they pick mark-read, loop through the indices of confirmation class only (never mark interview_invite, action_required, or recruiter_outreach as read — those need a human eye):

node scripts/outlook-triage.js --port "${OUTLOOK_PORT:-9224}" mark-read <idx>

After each action, re-run the search to get fresh indices (the script uses positional indexes into the current result list).

Rules (from root CLAUDE.md)

• Mark confirmation emails as read after logging. Leave interviews, assessments, and recruiter outreach unread.