Giving AI coding agents more tools doesn’t necessarily make them more efficient.
After using OpenCode and Claude Code heavily, I started focusing less on adding capabilities and more on controlling the information flowing into the agent.
That means:
- searching before reading entire files
- choosing between text, structural, and semantic search
- reducing noisy command output
- keeping sessions focused
- defining explicit stopping rules
I’ve documented the toolkit I currently use, including Headroom, RTK, Caveman, Serena, rg, fd, ast-grep, jq, yq, and others, along with how I define their responsibilities in AGENTS.md.
The principle behind all of it is simple:
Give the agent the smallest amount of information necessary to make the next correct decision.
Full post:
https://ammar-najjar.com/blog/local-ai-coding-toolkit/
Top comments (0)