Engineering insights on AI dev tooling, Claude Code skills, MCP servers, and the patterns that separate quality from slop.
The official MCP servers repo used to ship 20 reference implementations. Now it ships 7. The Postgres replacement has a known SQL-injection bypass still in use. Here's the migration map.
Read moreAndrej Karpathy's autoresearch repo has 80,700 stars and 630 lines of code. It was built to optimize ML training runs. The pattern — define a metric, define a constraint, let the model iterate overnight — generalizes to almost any task with a measurable outcome. We ported it to landing-page copy, STRIDE security audits, pricing-page A/B variants, and three other domains.
Read moreAnthropic shipped a new tokenizer with Claude Opus 4.7. The headline price ($5/$25 per million tokens) is unchanged from 4.6. The real bill on identical workloads can be up to 35% higher because the new tokenizer fragments the same input into more tokens. We did the math and the migration check.
Read moreSnyk's ToxicSkills report, published February 5 2026, found that 534 of ~4,000 public Claude Code skills had at least one critical security finding — malware distribution, prompt injection, or exposed secrets. We read the report, mapped what installs a malicious skill actually does, and built the install-time checklist.
Read moreFifteen months from a single tweet to an industry-wide replacement term. Here is the month-by-month timeline of vibe coding's rise, peak, and quiet absorption into agentic engineering.
Read moreLangfuse went to ClickHouse in January. Promptfoo went to OpenAI in March. The eval stack consolidated faster than anyone planned for. Here's the migration map.
Read moreGlama indexes 23,451 MCP servers. Trend Micro exposed 492 without auth or TLS. Snyk's portfolio scan found 66% with security issues. We worked through the corpus and named the 18 servers we would install on a machine we cared about.
Read moreVecta's February 2026 benchmark across 50 academic papers measured 69% retrieval accuracy for recursive 512-token chunking and 54% for semantic chunking. The default 'advanced' advice underperformed the naive baseline by 15 points. We walk through why, where semantic still wins, and the chunking config that actually shipped in our pipelines.
Read moreThe sixteen patterns that get a skill, agent, plugin, or MCP server immediately rejected from RuleSell — and why each is broken.
Read moreThree focused agents beat one generalist working three times as long. But only if you get the patterns right. Here's what works and what produces 10x bugs.
Read moreA well-prompted Sonnet beats a lazily-prompted Opus. The benchmarks agree — prompt engineering lifted GPT-4 accuracy by 50%, and a 9B model beat one 13x its size. Here's the data.
Read more67,000+ MCP servers exist across public registries. 38.7% require no authentication. 30 CVEs filed in 60 days. MCP is powerful, and the ecosystem is a mess.
Read moreA step-by-step tutorial with real code for building a Claude Code skill that triggers correctly, loads efficiently, and actually helps. Not abstract guidance — working examples.
Read moreA concrete walkthrough — from empty repo to published listing — for authors writing their first Claude Code skill.
Read moreStar ratings measure popularity. Download counts measure marketing. We measure quality directly — with automated signals and zero voting.
Read moreWe built a verified marketplace for AI dev assets because star ratings are broken and download counts measure marketing, not quality. Here's how RuleSell works and why.
Read more