jev-blocks

Design falling-blocks challenges in blocks.json to test Jev's decision limits under time pressure.

465|41|Updated Aug 4, 2026
One-click install
npx skills add https://github.com/autonomous-ai/openharness --skill jev-blocks-autonomous-ai
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: jev-blocks
Source: https://github.com/autonomous-ai/openharness/tree/main/store/agents/jev-blocks/skills/blocks
Command: npx skills add https://github.com/autonomous-ai/openharness --skill jev-blocks-autonomous-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? It helps you design and tune a falling-blocks puzzle challenge that reveals where an AI agent's real-time decision-making breaks down under gravity, speed, and time-cost constraints. ## Core Features & Use Cases - Challenge Configuration: Tune gravity, speedup, decision and move costs, garbage rows, piece weights, custom boards, and a style line in blocks.json. - Breaking-Point Analysis: Read .harness/verdict.json findings (timing, topout) to find the last setting that played clean and the first that failed. - Built-in Validation: Run the toolchain check script to catch out-of-range numbers, zero weights, or malformed board rows. - Use Case: Fix the seed, set speedup to 0, then raise gravity from 10 to 40 with garbage rows to pinpoint the rows-per-second rate where pieces start landing short of their target. ## Quick Start Ask the agent to design a Jev blocks challenge with gravity 25, speedup 0, and 8 garbage rows, then report where Jev starts missing placements.

Frequently Asked Questions about jev-blocks

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I design a falling-blocks challenge to test an AI agent?

Edit blocks.json to set gravity, speedup, decisionMs, moveMs, garbageRows, piece weights, an optional custom board, and a style line. Fix the seed, set speedup to 0, change one knob at a time, and let the game run about a minute per setting.

How do I find the breaking point of an AI playing a blocks puzzle?

Read .harness/verdict.json after each run and check the timing finding, which reports how many of the last 50 pieces landed short of target at a given rows-per-second rate. The last clean setting and the first failing one define the breaking point.

What do gravity, decisionMs, and moveMs mean in a blocks challenge?

Gravity is the fall speed in rows per second, decisionMs is the cost per decision, and moveMs is the cost per rotate or shift. A placement needs roughly decisionMs plus (moves + 1) times moveMs, and the piece locks wherever it is when time runs out.

Why does the blocks challenge pane show MOCK?

The pane shows MOCK when no API key is configured, meaning a fixed offline heuristic is making decisions instead of the live model. It proves the plumbing works but does not reflect real model performance, so always state which mode you measured.

How do I validate a blocks.json configuration?

Run node "$JEV_DSH/toolchain/check.mjs" to validate the configuration. It prints ok or one error line per problem, such as numbers out of range, all-zero weights, or board rows that are not exactly 10 characters of '.' and '#'.