Support Emails To Features

Incoming — Product

One-shot morning briefing: shows everything that landed on the user's plate without their initiation — new Jiras assigned to them by other people + email action items from Mail.app. Resolves "the user" from team.json (the PM running this).

This skill is a thin orchestrator. It delegates email handling to the same scripts used by /outlook-actions and Jira fetching to tools/jira-api.py. Both halves run in parallel.

Step 0: Bootstrap

Read silently:

1. team.json — to resolve the user's Atlassian accountId. The user is the PM running this (identify their entry in team.json); their atlassianId is the canonical lookup key. Don't hardcode the ID in this SKILL.md — always re-read from team.json.
2. .claude/skills/outlook-actions/owned-people.txt — reused for email classification (anyone-addressed-by-name → Action).
3. Parse $ARGUMENTS:
   • --since <Nd> — window for both Jira (created >= -Nd) and email (--since window). Default 7d.
   • --no-email — skip the email half.
   • --no-jira — skip the Jira half.
   • --mark-read — after the summary, prompt to mark surveyed unread emails as read (delegates to outlook-actions/scripts/mark-read.js).

Resolve repo root with the same pattern as /outlook-actions:

REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || pwd)"

Step 1: Fetch in Parallel

Run both fetches in parallel using two Bash tool calls in the same message.

Step 1a: Jira — newly assigned to user, reported by someone else

USER_ID="$(python3 -c 'import json; print(next(p["atlassianId"] for p in json.load(open("team.json")) if p.get("alias")=="MK"))')"
python3 tools/jira-api.py search "project = ${PROJECT_KEY} AND assignee = \"$USER_ID\" AND reporter != \"$USER_ID\" AND statusCategory != Done AND created >= -${SINCE_DAYS}d ORDER BY created DESC" --max-results 50 --fields summary,assignee,reporter,created,status,priority,labels

If the result is "No issues found" but you suspect that's wrong (e.g. you know there should be Jiras), run a sanity check:

curl -s -u "$ATLASSIAN_EMAIL:$ATLASSIAN_API_TOKEN" "https://$ATLASSIAN_SITE/rest/api/3/myself" | head -c 200

A 401 / "Client must be authenticated" response means the token is expired — surface this to the user with: "Your Atlassian API token has expired or is invalid. Refresh it at https://id.atlassian.com/manage-profile/security/api-tokens and update ATLASSIAN_API_TOKEN in .env." Then continue with the email half only.

Step 1b: Email — Mail.app unread inbox

osascript -l JavaScript "$REPO_ROOT/.claude/skills/outlook-actions/scripts/fetch-mail.js" "${SINCE_HOURS:-168}"

(168 hours = 7 days; convert from --since.)

Apply the same noise filter and classification logic as /outlook-actions Steps 2–5. Don't duplicate the rules — reference the SKILL.md if uncertain. Use owned-people.txt for the Product Manager/Khushali address-match override.

Step 2: Render the Combined Digest

Print this to terminal (markdown):

# 📥 Incoming for Product Manager — last <window>

## 🎫 New Jiras assigned to you (<N>)

| Key | Pri | Reporter | Created | Status | Summary |
|-----|-----|----------|---------|--------|---------|
| ${PROJECT_KEY}-XXXXX | High | Project Coordinator | May 12 | To Do | Add referral source filter to dashboard |
| ${PROJECT_KEY}-YYYYY | Med  | Jane Doe | May 11 | To Do | Spec subscription pause flow for v3.4 |
...

> If 0: "No new Jiras assigned to you in the last <window> — clean slate."

## 📧 Email action items (<M>)

| Pri | Sender | Subject | Received | Suggested action |
|-----|--------|---------|----------|------------------|
| P0  | Project Coordinator | Q2 sign-off needed | May 13, 02:27 IST | Reply with Meta-budget decision |
...

> If 0: "Inbox is clear — no email action items in the last <window>."

## ℹ️ Email FYI (<K>)

| Sender | Subject | Received |
|--------|---------|----------|
| ... | ... | ... |

Sort within each section:

• Jiras: by priority descending (Highest → Lowest), then created descending.
• Emails: by P0 → P1 → P2, then time_received descending within priority.

Step 3: Optional Mark-Read (only if --mark-read)

Skip if --mark-read was NOT passed.

Otherwise: prompt via AskUserQuestion to mark surveyed unread emails as read, same flow as /outlook-actions Step 7.5. Reuse .claude/skills/outlook-actions/scripts/mark-read.js.

For Jiras: never auto-acknowledge or transition — those are intent-bearing actions that need the user's explicit instruction. Only show them; don't touch.

Step 4: Final Summary

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Incoming — last <window>
Jiras assigned by others: <N> (highest: ${PROJECT_KEY}-XXXXX, <summary>)
Email actions: <M> (top: <subject>)
Email FYI: <K> • Filtered noise: <D>
Marked read: <R> emails
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Then offer ONE follow-up only if relevant:

• If high-priority Jira exists: "Want to /analyze-jira to plan it?"
• If high-priority email needs Jira: "Want me to convert action item #1 to a Jira via /jira?"
• Otherwise: silence is fine. Don't bombard with options.

What this skill does NOT do

• Does NOT include Jiras the user reported themselves (those aren't "incoming" — they're outgoing)
• Does NOT include Jiras already Done (statusCategory != Done filter)
• Does NOT include emails not in the unread + recent window
• Does NOT auto-create Jiras from emails — that's /outlook-actions per-row walk
• Does NOT push notifications, post to Slack, or alert anyone else
• Does NOT modify Jira state (no transitions, comments, or edits)
• Does NOT replace /plan-my-day — that's broader (PRs, meetings, dependencies). /incoming is the narrow "what arrived for me" subset

Edge Cases

• Atlassian token expired — surface clearly, continue with email-only output. Don't silently show zero Jiras.
• No unread email + no new Jiras — say so cheerfully ("Empty queue — go do focused work.") and exit.
• Many Jiras (>20) — show top 10 with a "+ N more" line. Most days have <10; if it's >20, something's wrong upstream and worth surfacing.
• Mail.app permission not granted — same error handling as /outlook-actions (point to System Settings → Privacy & Security → Automation).
• User invokes --since 30d — that's a "catch-up after vacation" pattern, fine. Cap at 90d.