You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A curated collection of agents, skills, and settings that supercharge Claude Code. The goal of this kit is to gather battle-tested subagents, reusable skills, and sensible default settings in one place so practitioners can drop them into their local Claude Code environment and immediately boost productivity.
Agents are role-focused subagents (language experts, reviewers, architects, etc.) that Claude Code can delegate work to.
Skills are self-contained capability packages (each with a SKILL.md) that teach Claude how to use a specific tool, library, or workflow.
Settings provide a ready-to-use settings.json with environment tuning, permission guardrails, and a status line.
🔧 Installation (user-level / global)
These steps install everything at the user level so the agents, skills, and settings are available across all your projects (Claude Code reads from ~/.claude/ on Linux/macOS and %USERPROFILE%\.claude\ on Windows).
Run the commands from the root of this repository.
Linux / macOS
Dependencies:node ≥ 18 (for hooks) and jq (for the status line). Both are optional if you skip hooks or the status line.
Requires jq (brew install jq / apt install jq). git is optional — the script degrades gracefully when not in a repo.
Windows (PowerShell)
Dependencies:node ≥ 18 (for hooks). Install via nodejs.org or:
winget install OpenJS.NodeJS
# 1. Create the target directories if they don't existNew-Item-ItemType Directory -Force `"$env:USERPROFILE\.claude\agents",`"$env:USERPROFILE\.claude\skills",`"$env:USERPROFILE\.claude\hooks\pre-tool-use",`"$env:USERPROFILE\.claude\hooks\post-tool-use",`"$env:USERPROFILE\.claude\hooks\notification"# 2. Copy all agentsCopy-Item agents\*.md "$env:USERPROFILE\.claude\agents\"# 3. Copy all skills (each skill is its own directory)Copy-Item skills\*"$env:USERPROFILE\.claude\skills\"-Recurse
# 4. Copy hooksCopy-Item hooks\pre-tool-use\*.js "$env:USERPROFILE\.claude\hooks\pre-tool-use\"Copy-Item hooks\post-tool-use\*.js "$env:USERPROFILE\.claude\hooks\post-tool-use\"Copy-Item hooks\notification\*.js "$env:USERPROFILE\.claude\hooks\notification\"# 5. Copy the status line scriptCopy-Item settings\statusline.ps1 "$env:USERPROFILE\.claude\statusline.ps1"# 6. Copy the platform settings and inject your username into path placeholders$settings=Get-Content settings\settings-powershell.json -Raw
$settings=$settings-replace'<YourName>',$env:USERNAME$settings|Set-Content"$env:USERPROFILE\.claude\settings.json"-Encoding utf8
Re-syncing / updates (recommended)
robocopy is the Windows equivalent of rsync — safe to re-run, only copies changed files:
If you have PowerShell 7 installed, open ~/.claude/settings.json after install and swap powershell for pwsh in the statusLine.command value. Windows Terminal is recommended for correct ANSI / UTF-8 rendering. No jq required — the script uses PowerShell's native JSON parser.
Heads up — settings. Copying settings.json overwrites your existing global Claude Code settings. If you already have a customized file, back it up first:
Note. The agents/ folder also contains a CLAUDE.md and an AGENTS-REFERENCE.md. These are reference/instruction docs, not agents — you can skip copying them, or copy them deliberately if you want their guidance.
After installing, restart Claude Code (or start a new session). The agents become available for delegation, and skills trigger automatically based on their descriptions.
🤖 Agents
133 subagents grouped by domain. Each links to its definition file.
Evidence-grounded answers from full-text research papers.
🧩 Skills
197 skills, each packaged as a directory containing a SKILL.md (and supporting scripts/templates). Skills trigger automatically based on their descriptions. Grouped below by domain.
statusLine — runs the platform-specific status line script (model, dir, git branch, context bar, cost, rate limits).
theme — dark.
model — sonnet[1m] (Sonnet with 1 million token context window).
Adjust model, theme, and the permission lists to match your own environment before relying on them.
A settings.local.json (machine-local permission overrides) is intentionally not tracked in this repo — it is git-ignored. Create your own at ~/.claude/settings.local.json if you need per-machine permission allowlists.
🪝 Hooks
The hooks/ directory contains Node.js scripts that Claude Code executes automatically at defined lifecycle events. Only the pre-tool-use hooks are wired into settings.json — the others (post-tool-use, notification) must be added manually if you want them.
Requirement: Node.js ≥ 18 must be on your PATH. The hooks use only built-in Node.js modules (fs, path, child_process) — no npm install needed.
Pre-Tool-Use hooks (active by default)
These run before Claude executes a tool call and can block the action entirely.
Access to sensitive files (.env, SSH keys, AWS/GCP/Azure credentials, keystores, .netrc, .npmrc, etc.) and bash commands that expose or exfiltrate secrets. Same SAFETY_LEVEL knob.
Post-Tool-Use hooks (optional)
Run after a tool call succeeds. Not wired into settings.json by default.
Sends a Slack message when Claude needs user input. Set CCH_SLA_WEBHOOK to your Slack Incoming Webhook URL.
Hook logs
All hooks write structured JSONL entries to ~/.claude/hooks-logs/YYYY-MM-DD.jsonl for audit and debugging.
💬 Prompting Tips
prompting_tips.md is a reference handbook for writing effective Claude Code prompts. It covers the 4 main prompt templates used in practice, with detailed field descriptions, filled-in examples, and patterns for invoking skills and subagents.
Section
What it covers
Template 1 — Development Task
[CONTEXT] / [TASK] / [CONSTRAINTS] / [ACCEPTANCE CRITERIA] structure for implementing or modifying code.
Template 2 — Bug Hunt
4-step protocol: Understand → Hypothesize → Reproduce → Fix. Each step is a separate prompt.
Template 3 — Test Generation
Happy-path + edge-case scenario structure that prevents Claude from forgetting error conditions.
Template 4 — User Story Generation
Job Story + Gherkin acceptance criteria for backlog grooming and feature specification.
Invoking Skills
How skill auto-invocation works, when it fires, and how to reinforce triggers.
Invoking Subagents
@name explicit call, proactive auto-invocation via description, and parallel Task-tool orchestration.
Combining Skills + Subagents + MCP
Real-world examples mixing all three mechanisms.
A pocket cheatsheet at the end summarises all four templates on a single printable card.
For the spec-driven development workflow (constitution → spec → plan → tasks before code), see speckit_cheatsheet.md — a quick reference for GitHub Spec Kit: install steps, CLI commands, the /speckit.* slash commands, and a worked example.
A curated collection of agents, skills, and settings that supercharge Claude Code. Drop them into their local Claude Code environment and immediately boost productivity.