I'm a CS Student in Placement Season. I Replaced My Inbox Anxiety With 80 Lines of Markdown.

This is a submission for the OpenClaw Challenge.

What I Built

I built inbox-triage — a personal email triage skill for OpenClaw. It's a single SKILL.md file, 80 lines of Markdown, no code, no API keys, no deployment. When I paste email subject lines or a screenshot of my Gmail into OpenClaw, it sorts them into four buckets:
🔴 URGENT — open now (OA deadlines, interview invites, placement cell mail)
🟡 REVIEW — today-ish (interview prep, company research, professor mail)
🟢 APPLIED — active applications awaiting response (Microsoft, Dbert Labs, Bluestock)
⚫ ARCHIVE — silently filed (LinkedIn noise, daily digests, noreply promos)
The problem it solves is specific: I'm a CS student in active placement season. My inbox gets 200+ mails a week and maybe 10 of them matter. I was missing OA deadlines while reading "5 people viewed your profile" notifications. The script version I built first (Gmail API + regex) broke every time a recruiter emailed from a domain I hadn't hardcoded. A skill handles the fuzzy cases a script can't.

How I Used OpenClaw

The whole project is one SKILL.md. OpenClaw's skills architecture does the heavy lifting — I just describe when the skill fires and what the rules are, in plain Markdown. Three pieces make it work:

The description frontmatter is the trigger. I wrote it to fire whenever I paste email info or ask "what should I open first." OpenClaw reads this and routes the request correctly without me invoking the skill explicitly.
Layered classification rules. The rules go URGENT → REVIEW → APPLIED → ARCHIVE, with an explicit tie-breaker: the more urgent bucket always wins. That single rule handles the hardest case — a [e-mail] email with subject "Interview scheduled Thursday 3 PM" is URGENT, not ARCHIVE, because the subject rule beats the sender rule.
A worked example in the skill itself. My first draft was rules-only and hit about 60% accuracy. Adding one fully-worked example with real sender names and expected output pushed it past 95%. Models pattern-match on examples harder than they follow abstract rules.

Demo

Input:

• From: [e-mail] — "Microsoft OA link — complete by 11 PM today"
• From: LinkedIn — "5 people viewed your profile this week"
• From: [e-mail] — "Re: Your application — next steps"
• From: Unstop — "Daily digest: 12 new hackathons"
• From: [e-mail] — "Interview scheduled: Thursday 3 PM"
• From: Medium Daily Digest — "Top stories in AI"

Output:
🔴 URGENT (open now)

• [e-mail] — Microsoft OA link — complete by 11 PM today — Placement cell + deadline today
• [e-mail] — Re: Your application — next steps — Applied company responding to your thread
• [e-mail] — Interview scheduled: Thursday 3 PM — Applied company + interview keyword beats noreply rule

⚫ ARCHIVE (3 items archived)

What I Learned

The hardest part of this build wasn't writing the rules. It was being specific enough about what "important" means to me that a model could apply it consistently.
"Placement cell emails matter" is obvious. But what about a LinkedIn notification that happens to mention Microsoft because I follow Microsoft? What about a noreply@ from a company I applied to last week? What about an Unstop email — is it a daily digest (archive) or a status update on something I applied to (surface)?
Every edge case forced me to articulate a preference I'd never actually put into words. By the time the skill was accurate, I understood my own email-reading heuristics better than before I started.
Two other things clicked:

A skill's description field is doing real work — it's not a comment for humans. Vague descriptions don't trigger reliably. Specific ones do.
Explicit output formats make skills reusable. The four-bucket structure with emojis means I can scan the output in two seconds. An open-ended format would've given me paragraphs — which is what I was trying to escape.

Skills are a forcing function for self-knowledge. I didn't expect that going in.

ClawCon Michigan

I didn't attend ClawCon Michigan — I was running Gmail triage from Baharampur, West Bengal, which is a few time zones and an ocean away. But the community around it is what made me want to ship something for this challenge instead of just reading the announcement.