Agent Config Starter — Development skill for Claude Code

agent-config-starter

[](https://github.com/damson/agent-config-starter/actions/workflows/ci.yml) [](LICENSE)

**Your agent instructions, managed like code.**

Hi! 👋 This is the on-ramp to [agent-config-harness](https://github.com/damson/agent-config-harness): a ready-to-run repo where your `CLAUDE.md` files are registered, structurally tested, and graded A to F by a rubric that notices when they get worse. The robots read these files before touching your code; this starter is how you stop feeding them folklore.

Everything here is pre-wired. The harness holds the machinery, vendored as a pinned submodule; every file you actually edit is config. Clone it, rename one example domain, and you are under management in five minutes.

Quickstart

gh repo create my-agent-config --template damson/agent-config-starter --private --clone
cd my-agent-config
git submodule update --init
just setup    # symlink configs and skills into ~/.claude
just check    # health-check every domain and link
just test     # secret lint + skill structure

⚠️ **Before `just setup`:** it points your global Claude config at this repo: `~/.claude/CLAUDE.md`, `~/.claude/preferences.md`, `~/.claude/statusline.sh` and the two identity files, plus your skills. An existing regular file there is moved aside to a timestamped `.bak` (and the run says so); a symlink owned by another config repo makes setup refuse rather than silently take over. Your old config survives either way, but read the output.

Scoring needs the Claude CLI and an API key. Everything else runs without one:

just eval            # score every domain on the rubric
just eval myproject  # one domain

What's in the box

A worked example of every moving part, each file carrying comments that say what to replace and w