Monitor Lizard - Using OpenClaw for the first time, I loved it!!!!!!

This is a submission for the OpenClaw Challenge.

What I Built

I'm from Fredericton, New Brunswick — a small province, one university, and not exactly a tech hub. I just finished my first co-op under UNB, and this summer I'm studying while working for a startup. But for my second internship this fall, I don't want to stay small. I want Toronto. I want Vancouver. I want to be somewhere that pushes me.

The problem is, from here, every job search feels the same. You open 15 career pages, scroll through hundreds of postings, copy-paste JDs into ChatGPT, try to keep track in a spreadsheet, and eventually lose motivation because you don't even know if you're applying to the right things.

I was browsing GitHub one evening and saw someone I follow had starred the CareerOps repo. I clicked through and thought — this is interesting. That same day, I had just made an account on DEV Community and saw the OpenClaw Challenge. It felt like the universe was telling me something.

So I thought: instead of manually hunting for my fall internship, what if I build something that does it for me? Not a chatbot. An actual agent that watches the market 24/7, scores every role it finds, and tells me exactly what I'm missing.

That's how monitor lizard was born.

It's a Discord-based autonomous agent built on OpenClaw that:

• Scans 60+ company portals daily — Greenhouse, Ashby, Lever APIs, plus 7 GitHub internship repos. It checks career pages so I don't have to.
• Scores every role A-F — across 10 dimensions like role fit, tech stack match, growth potential, work permit compatibility, and location. Every score gets a critic pass — a second review that challenges the first.
• Runs an overnight pipeline — 8 stages from 11 PM to 7 AM: scan, deduplicate, batch score, gap analysis, generate tailored CVs, draft follow-ups, sync interview stories, and send me a morning briefing. I wake up and the work is done.
• Generates tailored CVs — ATS-optimized PDFs customized to each job description. It doesn't fabricate skills — it reframes what I already have to match what they're looking for.
• Tracks everything in Notion — every role discovered, scored, applied to, rejected, or stale. One pipeline, one source of truth.
• Remembers context — uses ChromaDB vector memory so when I ask about a company, it remembers what it scored last week and adjusts.
• Does competitive intelligence — searches public profiles of past interns at a company, builds a picture of "who they typically hire," and tells me how I compare.

The name? Monitor lizards are patient predators. They don't chase — they watch, they wait, and when the moment is right, they strike. That's how you job search from a small province.

How I Used OpenClaw

OpenClaw is what makes the whole thing autonomous. It's the runtime layer—the Discord Gateway connection, the scheduler, and the skill dispatcher.

The bot lives on Discord. I DM it commands like /score, /compare, /scan, /cv, /digest, or just talk to it naturally ("find me ML jobs in Toronto"). OpenClaw handles the WebSocket connection to Discord, receives my messages, and routes them through my interceptor pipeline before dispatching to the right skill.

12 commands, all wired through OpenClaw:

Command	What It Does
/score [JD]	Score a role A-F across 10 dimensions
/compare [URL]	3-layer gap analysis against a job posting
/cv [URL]	Generate a tailored ATS-optimized PDF
/scan	Crawl 60+ portals for new listings
/career-ops	Full pipeline: scan → score → log → notify
/lizard intel	Competitive intelligence on a company
/noctis	Trigger 8-stage overnight pipeline
/followup [company]	Draft follow-up for stale applications
/digest	Evening pipeline summary
/pipeline	Show Notion pipeline status
/prep [company]	Interview prep with STAR stories
/outcome [company] [result]	Log outcome for calibration

The scheduler runs everything on autopilot. OpenClaw's heartbeat scheduler fires events on a cron-like schedule — career-ops at 8 AM, digest at 8 PM, noctis at 11 PM, follow-up checks every 6 hours. Each event either runs in-process (fast tasks like scoring) or spawns a child process (long tasks like the overnight pipeline) so the bot never drops its Discord connection.

The agent can actually do things. When I send a general message, Claude gets 8 real tools — search jobs via Brave, add roles to my Notion pipeline, evaluate a JD, run competitive analysis, search the web via Tavily. It doesn't just talk about jobs — it finds them, scores them, and logs them.

Smart model routing saves money. Not every task needs Claude Sonnet. The scanner runs on Haiku ($0.01/call). Evaluations use Sonnet ($0.15/call). This brings the cost down to about $17/month instead of $45 — which matters when you're a student.

The interceptor catches mistakes before they happen. Every message passes through 4 stages: classify the command, inject relevant memory from past evaluations, check safety gates (work permit flags, rate limits, duplicate detection), and route to the right model. The gates are code, not prompts — an LLM can forget a rule, but if (!confirmed) return BLOCKED can't be hallucinated away.

Demo

monitor lizard — noctis briefing — 2026-04-23

scanned: 63 portals. 7 new roles.
scored: 7 roles. 2 rated A, 3 rated B.
compared: 2 A-tier roles triaged.
cvs: 2 tailored PDFs generated.
followups: 1 draft queued (Shopify — 9 days stale).

top picks:
  [A] Wealthsimple — ML Platform Intern — 4.6/5.0
  [A] Cohere — Applied AI Co-op — 4.4/5.0
  [B] RBC — Data Engineering Intern — 3.9/5.0

The project structure:

scripts/
├── openclaw/       # Runtime — gateway, scheduler, dispatcher, agent (9 files)
├── interceptor/    # Message pipeline — classify, gate, route (6 files)
├── skills/         # Intelligence — evaluator, compare, cv, scan, noctis (9 files)
��── clients/        # APIs — Discord, Notion, Brave, Tavily, ChromaDB (5 files)
├── utils/          # Helpers — PDF gen, dedup, liveness checks (11 files)
└── test-all.mjs    # 172 tests, 0 failures

What I Learned

The honest version of how this started: I wasn't planning to build anything. I was just a student in Fredericton scrolling GitHub after dinner. Someone I follow had starred the CareerOps link. I clicked it, read about OpenClaw, and it clicked this is a framework I could build something real on. That same day I'd signed up on DEV and the OpenClaw Challenge was right there on the page. So I just started building. No grand plan. I need a fall internship, and I'm tired of doing this manually.

The feature I'm most excited about building next is what I'm calling the intern compare mode. Right now, the compare skill does gap analysis against a job description it tells me what skills I'm missing. But what I really want is a pipeline that goes further: find actual people on LinkedIn and GitHub who work at the company, compare my resume against theirs, and tell me exactly what skills they have that landed them the job that I don't have yet. Not generic advice like "learn Kubernetes." Specific, data-driven answers like "3 out of 4 ML interns at Cohere had a public model-serving project on GitHub you don't." That's the kind of signal that actually changes what you work on next. I've already started building parts of this into the lizard intel skill and I want to make it the core of how the agent helps me prepare.

The thing I'll take with me regardless: building this forced me to think about my career like a system. What are my real strengths? Where are the actual gaps? What does the data say about who gets hired at the places I want to work? The monitor lizard doesn't care about my feelings. It just shows me the score. And honestly, that's what I needed — not motivation, but clarity.

ClawCon Michigan

I did not attend ClawCon Michigan but Hopefully one day i want to