dominiks-skills

Dominik's Skills

My personal coding-agent skills, organized in a Pocock-style bucket layout. One installer, any harness — Claude Code, Pi, Codex, Gemini CLI, Cursor, and the shared ~/.agents/skills/ location. Both my originals and the upstream Pocock skills, all aligned to my conventions (gh-only issues, parent + sub-issue decomposition, mandatory Test Strategy, Pocock-style TDD).

Adapted from Matt Pocock's mattpocock/skills — the structure, several skills, and the TDD discipline are his.

Quickstart

git clone https://github.com/dominikdorfstetter/skills.git
cd skills
./scripts/install.sh

The installer walks you through two steps:

1. Select harness(es) — Claude Code, Pi, Codex, Gemini CLI, Cursor, or the shared ~/.agents/skills/ location. Pick as many as you want.
2. Select buckets — which skill categories to install. Defaults: engineering, productivity, misc are Y; personal, deprecated, and new buckets are N.

Each bucket is discovered automatically — anything under skills/ with at least one SKILL.md shows up.

Non-interactive (CI / scripting)

# Install default buckets into Claude Code only (same as old link-skills.sh --yes)
./scripts/install.sh --harness claude --yes

# Everything, every harness
./scripts/install.sh --harness all --all

# Specific harnesses, specific buckets
./scripts/install.sh --harness claude,pi --buckets engineering,productivity

# Shared location only, default buckets
./scripts/install.sh --harness agents --yes

# Preview
./scripts/install.sh --harness all --all --dry-run

# Clean stale symlinks first
./scripts/install.sh --harness claude --clean --yes

./scripts/install.sh --help

Backward compatibility

./scripts/link-skills.sh still works — it's a thin wrapper that calls install.sh --harness claude with the same flags as before.

Run from anywhere — setup-skills shell command (zsh / bash)

For a setup-skills command available in any directory, append this to ~/.zshrc (or ~/.bashrc):

# >>> dominiks-skills installer (managed) >>>
setup-skills() {
  local script="$HOME/repos/skills/dominiks-skills/scripts/install.sh"
  if [[ ! -x "$script" ]]; then
    echo "setup-skills: installer not found at $script" >&2
    return 1
  fi
  "$script" "$@"
}
# <<< dominiks-skills installer (managed) <<<

Reload (source ~/.zshrc) or open a new shell, then:

setup-skills                         # interactive (buckets + harnesses)
setup-skills --harness claude --yes  # Claude-only, default buckets
setup-skills --harness all --all     # everything everywhere
setup-skills --dry-run --all --harness all
setup-skills --buckets engineering,personal --harness claude,pi

Adjust the $HOME/repos/skills/dominiks-skills/... path if you cloned the repo elsewhere.

Conventions

Issue-producing skills (bug-report, feature-pm, architect-review, security-audit, resolve-issue, to-issues, to-prd, triage, improve-codebase-architecture) follow a consistent contract:

1. All issue I/O goes through gh — gh issue view, gh issue list --search, gh issue create, gh issue edit, gh issue comment. Never assume a different tracker.
2. Non-trivial work is a parent epic + sub-issues. Parent body lists - [ ] #N checkboxes; each sub has Parent: #N. Parent and sub link both ways. to-issues is the standard decomposer.
3. Every issue ships a Test Strategy table with a named tracer-bullet test, plus rows for happy path, permissions, validation, and at least one edge case.
4. Implementation is always TDD — Pocock-style red-green-refactor in vertical slices. One test → minimal impl → repeat. Never write all tests first. See the bundled tdd skill.
5. Skills stay language- and framework-agnostic. Even scaffolders describe artifacts/layers (schema, repo, handler, routing) — never specific frameworks.

Reference

Engineering