Articles on AI Agents & Tools

Complete architecture breakdown — smart router, 16 mode handlers, 18 free APIs, dynamic source weighting, citation density, and the iterative ratchet quality process.

The extraction → generation → score → improve pattern that took quality from 56 to 94/A. Model routing decisions, the thinking-model gotcha, rate limiting, and why vague descriptions kill quality at step 1.

CLAUDE.md is the system prompt file that tells Claude who it is, what it can do, and what it must never do. This guide covers structure, iron laws, communication sections, and what separates a B-grade from an A-grade system prompt.

OpenRouter's free tier gives you access to 28 models with up to 1M token context windows — no credit card required. Here's what's actually buildable, which models are reliable, and the gotchas that aren't documented anywhere.

Empirical token saturation research across 20 skills: at what token budget does quality plateau? The routing table that sends simple tasks to GLM-5 and complex ones to Claude.

A 3-way blind eval: vocabulary blacklist vs generic rewrite vs full 8-pass humanizer skill. The gap isn't in word choice — it's in rhythm, sentence variance, and structural naturalness. The data behind the AI Text Humanizer.

Most SKILL.md files fail quietly. An 8-point quality rubric, token saturation data from 20+ skills, and what a linter actually checks: vague triggers, absent iron laws, token bloat.

We tested 20 skills against real prompts, measuring completeness, quality, and token efficiency. Four failure modes dominate bad skills. Here's what separates a 94/A SKILL.md from a 40/F one — with data.

Vocabulary removal scored 31/60. LLM rewrite scored 31/60. 8-pass structural revision scored 55/60. The gap isn't in word choice — it's in rhythm, CoV, and transition density. Blind eval data and a fix process.

Implementation notes and three evals. Convergence quality: 4/10 → 9/10. Contradiction detection: 7/10 → 10/10. Narrow factual queries: tie. The tie was the most informative result.

The complete guide to free tools for building and running AI agents: free LLM APIs, memory systems, web search, MCP servers, code execution, and monitoring.

Practical guide to effective system prompts and CLAUDE.md files. Covers structure, iron laws, permission modes, memory, and what separates working instructions from noise.

How the SKILL.md file format lets AI agents acquire new capabilities dynamically. What SKILL.md is, how skills trigger, and best practices for building reliable agent skills.

Anatomy of a good CLAUDE.md: iron laws, permission modes for headless agents, memory patterns, and the common mistakes that make Claude ignore your instructions. Includes a free generator.

Complete guide to the SKILL.md format — anatomy, trigger types, common mistakes, and best practices. Includes a full annotated example and free generator.

Test your system prompts and CLAUDE.md files before deploying. Adversarial testing strategies, the five tests every prompt needs, common failure patterns, and a free automated instruction tester.

Ranked comparison of free AI humanizer tools tested against real detectors. Which ones change structure vs. just swap synonyms, and why multi-pass revision beats single-pass rewrites.

How AI content detectors actually work, why they fail on non-native speakers and edited text, the real signals of AI writing, and how to audit text manually. Pattern-based approach beats single-score detectors.

A free AI coding assistant with no signup, no API key, no rate limits. Supports Python, JavaScript, Rust, Go and more — powered by smart model routing across free LLM APIs.

Generate regex patterns from plain English descriptions. Free AI-powered regex builder with copy-as support for JavaScript, Python, Java, Go, and PHP. No signup required.