01 / #ai · 8 posts
2026
-
· protocols · 16 min
A2A Is for Strangers: When to Use A2A and When It's Overhead
When to use A2A: at a boundary where a standard contract pays off. Inside a tightly coupled fleet, it's usually overhead. A Docker test against MCP and NATS.
-
· agents · 16 min
When Not to Use RAG: Log Your Misses First
Most empty searches in my agent's memory were plumbing, not meaning. When not to use RAG: a week of logs, and bge-m3 against an LLM judge.
-
· claude-code · 7 min
A Claude Code Mod So I Never Miss My Agent's Questions
A Claude Code mod that keeps your agent's open questions above the prompt: about 250 lines, written by Claude itself, and how to build your own.
-
· claude-code · 19 min
Claude Code Hooks: The Deterministic Layer You're Not Using Yet
Claude Code hooks guide: all 33 lifecycle events, exit code 2 blocking, PreToolUse examples, the stop_hook_active trap, and the patterns worth shipping.
-
· claude-code · 15 min
Claude Code SKILL.md Frontmatter: Every Field Explained
Claude Code SKILL.md frontmatter reference: all 20 fields, the context: fork and agent pair, body templating, and the bugs you'll hit in practice.
-
· claude-code · 9 min
ForgeCode vs Claude Code: What Happened to the 81.8% Benchmark
ForgeCode vs Claude Code: how ForgeCode's 81.8% Terminal-Bench score got rescored, where both agents stand on Terminal-Bench 4.0, and which CLI to pick now.
-
· claude-code · 10 min
Claude Code Source Leak: What It Revealed and What Shipped Since
The Claude Code source leak, six months on: what the 2.1.88 source map exposed, which hidden features shipped, which died, and how the agent really works.