wrhn

Recent submissions

Comment · 10 months ago
I built a harness that lets Claude Code play Clash Royale autonomously through screen capture and input automation. The AI makes decisions from raw screenshots and executes them via shell scripts that click screen coordinates. The core challenge was latency. Each cycle (screenshot → vision analysis → decision → tool call) takes ~7 seconds. Far too slow for a real-time game. My solution: spawn 3 parallel agents playing the same match simultaneously. Their natural stagger produces a card play every 2-3 seconds. Brute force, but effective. Architecture is a Commander agent managing menus and game flow, spawning Player sub-agents for each match. The coordinate system maps UI elements and a placement grid so agents can call play_card.sh B2 giant and have it translate to pixel coordinates. Results: 1000+ trophies (Arena 5). Below 1000 you're mostly playing bots - crossing that threshold means real players with better cards and actual strategy. Latency becomes punishing when humans can exploit slow reactions. Interesting findings:

Claude researched its own deck strategy and wrote the sub-agent gameplay prompts The game shows "WINNER!" on every result screen including losses - had to teach it to verify via trophy delta Added an auto-opener (first card at 5 sec) because Claude was too slow to establish early tempo

Streamed 12+ hours on Twitch. Claude played 5 consecutive games with minimal human intervention (just closing pop-ups and opening chests). To my knowledge this is the first harness for frontier AI playing Clash Royale. Potentially interesting as a benchmark - tests vision, real-time decision making, resource management, and multi-agent coordination.

Comment · 1 year ago
This is really cool. Really interested in your MCP workaround - I have three future agents blocked by Issue #7296. Would you be willing to share more details about your GLM-4.6 setup? Specifically how you're doing "API (not true MCP behaviour)" to get sub-agents accessing those tools? house-mcp would be super useful for me
Comment · 1 year ago
Hey HN,

I've been thinking about the actual use-case for sub agents and how they could solve problems I run into doing day-to-day dev. I have 6 use cases but built three of them so far and have some good results testing so far.

When Claude searches my codebase or analyzes a big git diff, my context window fills up with thousands of tokens of crap that the master instance almost never needs. After a couple of these operations, I'm basically starting over because there's no room left for actual work.

I built specialized sub-agents that run in their own contexts and return condensed summaries instead of dumping everything into my main conversation.

house-research: Searches codebases/docs and returns condensed findings with file:line references instead of dumping full file contents into your context. house-git: Analyzes diffs and commits, returns impact summaries grouped by severity instead of raw git output.

house-bash: Runs commands and parses output for key info (pass/fail counts, relevant errors) instead of flooding your context with full logs.

Three I haven't done yet: house-mcp (mcp), house-vision(screenshots), house-data(data queries)

Results: 90-95% token reduction on heavy operations. My master instance stays focused on implementation instead of getting clogged with search results and logs.

What operations eat your context that I should build agents for?

One time install with CC: Clone https://github.com/houseworthe/house-agents. Detect my OS and copy the .claude directory to my current project. Verify all three agent files exist in .claude/agents/ directory. List available sub-agents to confirm they load. Test house-research by finding TODO comments.

Live from the official Hacker News API

A modern reader for Hacker News · SvelteKit + Skeleton