Claude Code Configuration, Skills, Releases, and Cost Evidence

An ECC plugin evaluation measured about 13.7k tokens of skill and agent descriptions and inspected, but did not run, 27 hooks; a skill-effectiveness method advocates ablation, repeated runs, and graders. New AGENTS.md guidance highlights version-specific precedence: CLAUDE.local.md can block AGENTS.md loading, while a configuration option can load both. SessionStart hook examples and v2.1.318–2.1.319 notes on deny-instead-of-prompting, background MCP tasks, and security fixes remain useful but should be verified against the installed version.