dotagents

My shared Claude Code configuration: skills, slash commands, subagents, and hooks I reuse across projects, plus a small house style. One repo, symlinked into the global ~/.claude so it applies everywhere.

What is in here

.claude/
  CLAUDE.md      global house style (short, universal rules)
  skills/        on-demand capabilities and house-style specs
  commands/      custom slash commands (/<name>)
  agents/        custom subagents
  hooks/         PreToolUse lint scripts (python3)
settings.global.json   house preferences merged into ~/.claude/settings.json
install.sh             link components + merge settings; also `enable-hooks`
tests/                 bash tests for the hooks + a repo-wide ASCII self-lint

Global vs opt-in

• Global (always available): skills/, commands/, agents/, and the house CLAUDE.md. These are advisory or on-demand, so applying them everywhere is harmless.
• Opt-in per project: the lint hooks. Their scripts are linked globally so they have a stable path ($HOME/.claude/hooks/...), but they only fire in a project whose .claude/settings.json activates them. This keeps other people's repos (and ones that legitimately need unicode or different commit styles) free of the enforcement.

Install

git clone https://github.com/timcane/dotagents
cd dotagents
./install.sh

install.sh is idempotent. It:

• symlinks skills/, commands/, agents/, hooks/, and CLAUDE.md into ~/.claude (override the location with CLAUDE_CONFIG_DIR);
• backs up anything already at those paths to ~/.claude/backups/dotagents-<ts>/ before replacing it;
• merges settings.global.json into ~/.claude/settings.json (preserving your existing keys such as theme and enabledPlugins).

Enable the lint hooks in a project

From inside the project you want to lint:

~/path/to/dotagents/install.sh enable-hooks

This merges the opt-in hook block into that project's .claude/settings.json, pointing at the globally linked scripts. Re-running is safe (it will not add duplicates).

Hooks

• ascii-only-lint.py - blocks Write/Edit/MultiEdit content containing non-ASCII bytes. Pairs with the ascii-only skill.
• git-message-lint.py - blocks git commit / gh pr create|edit messages that break the house style (non-Conventional subject, over 72 chars, prose-paragraph bodies, AI attribution footer). Pairs with the commit-style and pr-style skills.

House style

The global CLAUDE.md carries the short, universal defaults (Conventional Commits, terse PR descriptions, no AI attribution footer). The detailed specs and examples live in the commit-style, pr-style, compose-style, and ascii-only skills. A repo's own conventions override these where they conflict; ASCII-only is opt-in, never forced globally.

Tests

bash tests/run.sh

Runs the hook tests and the repo-wide ASCII self-lint. CI (GitHub Actions) runs the same on every push and pull request.

Credits

The grill-me skill is adapted from Matt Pocock's version; see .claude/skills/grill-me/CREDITS.md.

License

MIT - see LICENSE.