dotclaude-skills
zommuter/dotclaude-skills
Public Claude Code skills — meeting, and more
Instalação
npx convez-ai zommuter/dotclaude-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
dotclaude-skills
Public Claude Code skills by @zommuter.
Skills
| Skill | Description |
|---|---|
| meeting | Structured design meetings with named personas (Archie, Riku, Petra); --cross mode scans all registered projects |
| git-diary-workflow | Commits, pushes, and appends a diary entry after every substantive prompt |
| todo-update | Updates TODO.md after every substantive prompt — mandatory alongside git-diary-workflow |
| relay | The reviewer/executor relay: autonomous pool (bare /relay), handoff, anti-gaming review, reconcile (human-invoked parked-orphan disposal), and human modes, plus the lean executor contract loaded by /relay executor (Sonnet sessions in relay-managed repos) |
| projects | Personal project dashboard — recent activity, open task counts, prioritization |
(fables-turn / fables-executor remain as deprecated alias stubs that forward to
/relay and /relay executor.)
The relay
A strong reviewer model prepares repos (docs, roadmap, failing tests as the
spec) and audits finished work; cheap executor sessions grind the well-specified
items in between. docs/relay.md is the
user-facing guide: run modes, what every artifact means (ROADMAP.md,
RELAY_LOG.md, REVIEW_ME.md, fable-ckpt-* tags, relay.toml, RELAY_STATUS.md),
and what the human does between turns.
Statusline & tools
- statusline/ — quota/cost/model statusline with pricing-window
indicator and session token total (
statusline-command.sh, reads JSON on stdin). - tools/ —
allowlist.py(settings.json Bash-allowlist generator) andctx-budget.sh(advisory SKILL.md token-budget audit).
Hooks
Event hook scripts for Claude Code. See hooks/README.md for settings.json registration snippets.
| Hook | Description |
|---|---|
| meeting-cost-logger.sh | Stop hook: logs per-session turn/token counts and whether a meeting note was written |
| parallel-edit-detector.py | Stop hook: detects committed file changes not explained by the session's tool calls |
| notify-hook.linux-x11.sh | Notification hook: desktop notification for permission prompts (XFCE/X11) |
Install
git clone https://github.com/zommuter/dotclaude-skills.git ~/src/dotclaude-skills
cd ~/src/dotclaude-skills
make install # all skills, hooks, and allowlist entries
make install-meeting # one skill
make install-hooks # hooks only
make help # list available targets
make install calls make install-allowlist which merges the required Bash(...) entries into
~/.claude/settings.json (backup → settings.json.bak). The merge is idempotent: re-running
make install-allowlist adds nothing if all entries are already present. Use
make print-allowlist for a read-only preview (shows + for missing, = for existing).
See each skill's README.md for skill-specific notes (local-only files, etc.).
See hooks/README.md for hook prerequisites and additional settings.json snippets.
Publishing pattern
Each skill directory contains the public-safe spec files only. Personal accumulator files (discoveries.md, user-profile.md) are kept as local-only files in ~/.claude/skills/<skill>/ and never committed here.
The recommended install uses per-file symlinks (P2 pattern): ~/.claude/skills/<skill>/$f → ~/src/dotclaude-skills/<skill>/$f for each spec file. This means the live skill is the published version — no manual sync, no drift.
See each skill's README.md for install instructions.
Licence
MIT — do what you want, attribution appreciated.