01THE PRESSURE
Problem
Agent coding rules often accumulate as untested opinions rather than responses to measured failure.
CASE STUDY / Agentic Development Toolkit
claude-kit
Guardrails should come from observed failure patterns, not intuition alone.
Node.jsPythonPlaywright
Open-source agent skills and guardrails scoped from counted telemetry rather than opinion.
WHAT THIS PROVES
Developer tooling / Open source / Telemetry-scoped rules / MIT licensed
CLAUDE_KIT / GUARDRAIL_LOOPOPEN SOURCE / MIT
STAGE_01
Observe repeated agent behavior without guessing.
Interactive system walkthrough · simulated frontend data · not live telemetry
02THE SYSTEM
Approach
Use counted telemetry to identify recurring failure modes and scope focused skills and guardrails.
03THE MEASURE
Experiments
Observe behavior, count failures, then keep only rules that answer an evidenced pattern.
04THE BREAK
Failure mode
Overbroad rules can make agents less effective while appearing safer.
05THE JUDGMENT
Engineering decision
Prefer small, evidence-backed constraints over sprawling instruction sets.
06THE EVIDENCE
Measured result
Reusable agent tooling for other developers, released under MIT.