jhostalek-skills

dotclaude

skills for agentic coding tools. extremely opinionated. updated (almost) daily.

heads up: this is a global ~/.claude configuration repo — skills, hooks, and settings that apply across all projects. project-specific instructions (CLAUDE.md, TOOLS.md, skills) live in individual repositories.

build

start here. sparring pressure-tests a proposal, design explores alternatives, and plan turns the selected direction into an implementation-ready design document.

skill
sparring	adversarial pressure-test a proposal before committing
design	divergent exploration with independent reasoning agents
plan	reviewed implementation-ready design documents

quality

before you ship. review what you built, refactor what's messy, audit what's bloated or stale.

skill
judge	independent expert review before accepting work
audit	full audit sweep — sequences every focused audit below
audit-necessity	question whether code should exist, cut maintenance cost
audit-structure	fix misplaced files, god-modules, leaky boundaries, layering violations
audit-patterns	unify divergent implementations, kill reimplementations
audit-correctness	hunt logic bugs — off-by-one, inverted conditions, boundary cases
audit-error-handling	hunt swallowed errors, missing awaits, unguarded entry points, resource leaks
audit-logs	fix missing observability, INFO bloat, wrong-level messages
audit-perf	catch N+1, overfetching, blocking hot paths, unbounded growth
audit-security	injection, auth/authz gaps, secrets, weak crypto, OWASP
audit-tests	find redundant/weak tests, cover uncovered code
audit-complexity	maximize LOC reduction
audit-comments	strip stale or bloated comments and docstrings
ux	UI evaluation across Nielsen's heuristics

ship

git workflow from branch to PR to merge.

skill
commit	conventional commits + push
pr	create or update a reviewable PR
merge	merge branch into current, resolve conflicts
rebase	rebase branch onto target, resolve conflicts
timesheet	monthly work summary from git

improve

sharpen prompts, skills, and communication.

skill
prompt	create and refine LLM system prompts
transformer	rewrite skills for reasoning model performance
design-refiner	turn UI build requests into expert design briefs
writing-identity	rewrite communication in the user's natural voice

hooks

hook
approve-piped-bash	auto-approves piped commands when every segment is already in your allowlist (#1271)
auto-approve-all	auto-approves all tool permission requests
detect-skill-invocation	type /skillname in a prompt, get the skill injected into context (#19729)

configuration

skills work out of the box, but some features require settings.json entries:

setting	why
env.CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS: "1"	required for skills that spawn agent teams (judge, design, audit)
permissions.additionalDirectories: ["~/.claude/skills"]	lets teammates read skill files — needed if you symlink skills into ~/.claude/skills/

hook scripts ship in hooks/ but must be wired in settings.json under the hooks key to take effect — see Claude Code docs on hooks.

install

Codex plugin

Add the marketplace and install the plugin:

codex plugin marketplace add JHostalek/dotclaude
codex plugin add jhostalek-skills@jhostalek

Claude Code plugin

Add the marketplace and install the plugin:

/plugin marketplace add JHostalek/dotclaude
/plugin install jhostalek-skills@jhostalek

Skills are versioned with the plugin and invoked as /jhostalek-skills:<skill-name>. The plugin intentionally contains skills only; configure hooks separately if you want to use them.

Editable standalone skills

# clone and symlink
git clone [email protected]:JHostalek/dotclaude.git ~/dotclaude
ln -s ~/dotclaude/skills/* ~/.claude/skills/

# or just copy what you need
cp -r skills/pr ~/.claude/skills/

license

CC0 1.0 — public domain.