A reproducible, verifiable Claude Code setup — skills, agents, hooks and provider switching, installed with one command.
Getting started · Configuration · Providers · Security · Troubleshooting
Claude Code is the upstream harness. This toolkit keeps it that way, then layers on top: curated skills, agents, hooks, MCP starters, provider switching, and the checks that tell you what is actually active.
- Install only what you want. Four plugins via the Claude Code marketplace, or the full bootstrap in one command on Linux, macOS, native Windows or WSL.
- Context cost is measured. The skill listing is CI-enforced at 2,048 chars — half of what a 200k-token context allows. Without that cap, Claude silently drops skills past its 2% budget.
- Every skill is pinned to a reviewed commit SHA, licence-checked, and recorded with a content digest.
- Updates never destroy your work. Fast-forward only; a conflicting update refuses and tells you rather than resolving silently.
- Upstream updates arrive as reviewed PRs. Routine changes auto-merge; anything with capability surface (hooks, agents, credentials, network access) waits for a human.
- Provider switching without pretending. Codex uses the official CLI/App Server path; tokens stay with their native owner. Gemini/Vertex via AI Studio key or Vertex credentials. Personal Antigravity consumer OAuth is never proxied into Claude Code.
- No telemetry, no account, no service.
/plugin marketplace add furkankoykiran/.claude
/plugin install fk-gh-flow@fk-toolkit
| Plugin | What it does | Listing cost |
|---|---|---|
fk-gh-flow |
Find issues, solve into PRs, follow up on review, write human-sounding comments | 746 chars |
fk-writing-kit |
Strip AI tells from drafts; build posts from a chat, URL or GitHub profile | 819 chars |
fk-eng-agents |
researcher · planner · code-reviewer · debugger subagents | 0 |
fk-toolkit-ops |
Manage updates, MCP setup, Codex-native ImageGen boundaries | 233 chars |
Plugins carry no hooks and no executables.
Linux · macOS · WSL · Git Bash
curl -fsSL https://raw.githubusercontent.com/furkankoykiran/.claude/main/install.sh | bashWindows (native PowerShell)
irm https://raw.githubusercontent.com/furkankoykiran/.claude/main/install.ps1 | iexRe-running is safe. Verify the install:
bun install --frozen-lockfile
bun run catalog:check
fkt doctorSwitch provider and verify:
# Anthropic (default)
ccs anthropic
# Codex (experimental; see provider matrix for live-tested paths)
fkt setup --yes --provider codex --model gpt-5.5 --effort medium
ccs login codex
ccs codex
ccs status
# Official local AGY CLI (experimental; sign in with agy itself)
# No Gemini API key; optional AGY_GATEWAY_MODEL comes from: agy models
ccs agy
# Gemini Developer API / Vertex (separate credentials and model required)
# Set GEMINI_API_KEY and GEMINI_GATEWAY_MODEL, or configure supported Vertex.
ccs geminiccs google and ccs antigravity are non-switching migration errors. Failed Google preflight leaves the active provider unchanged.
See Provider capability matrix for tested boundaries. Codex, AGY and Gemini are experimental — see Status and limitations.
fkt check # is there an update?
fkt update # fast-forward, then run migrationsflowchart LR
U["developer"] --> CC["Claude Code upstream CLI"]
CC --> S["skills, agents, hooks"]
CC --> MCP["Claude-owned MCP servers"]
P{"active provider"} --> A["Anthropic"]
P --> C["Codex loopback gateway"]
P --> G["Google gateway adapters"]
P --> O["other Anthropic-compatible providers"]
CC --> P
C --> APP["official Codex App Server"]
APP --> CHATGPT["ChatGPT/Codex entitlement"]
G --> AGY["official local AGY CLI: own login/session"]
G --> GEMINI["Developer API / Vertex: own credentials"]
C -. "no token copying" .-> MCP
G -. "Claude owns tool execution" .-> MCP
Claude Code remains the executable you run. The toolkit changes configuration, not the
harness. Provider gateways are pinned submodules at gateways/.
flowchart LR
M["skills-sources.toml"] --> RES
G["git sources"] --> RES["resolver<br/>pin, fetch, verify licence"]
R["runtime sources"] -. metadata only .-> RES
RES --> L["skills-source.lock.json<br/>catalog/cache"]
L --> GEN["generator"]
GEN --> OUT["catalog/generated/"]
OUT --> PR["automation pull request"]
PR --> GATE{"policy gate"}
GATE -- routine --> AM["squash auto-merge"]
GATE -- capability surface --> HUMAN["held for human review"]
AM --> REL["tagged release<br/>with SHA256SUMS"]
Every skill is pinned to a reviewed SHA, licence-checked, and recorded with a content
digest. Two consecutive bun run catalog:generate runs produce byte-identical output.
| Provider | Status | Notes |
|---|---|---|
| Anthropic | Stable default | Full Claude Code feature parity |
| Codex | Experimental | GPT-6.1 Sol text, Claude-owned Bash, timeout recovery, and upstream Auto tested; see provider matrix |
| Google / Gemini | Experimental | Requires official AI Studio key or Vertex credentials; no consumer OAuth |
| NVIDIA, Z.ai, DeepSeek, Kimi, MiniMax, OpenRouter | Template | Provider-dependent; not live-tested in this repository |
.claude-plugin/ generated marketplace manifest
AGENTS.md shared cross-agent repository instructions
bin/ executables on PATH (cc-provider, fkt)
bun.lock dependency lock
catalog/ catalog toolchain (src, tests, cache, generated)
CLAUDE.md repository instructions for Claude Code
CODE_OF_CONDUCT.md community standards
config.json.example template seeded to config.json on install
CONTRIBUTING.md contribution guide
docs/ documentation — start at docs/README.md
gateways/ pinned runtime submodules (Codex and Gemini gateways)
hooks/ git and Claude Code hooks
install.ps1 installer for native Windows — public URL, do not move
install.sh installer for Linux/macOS/WSL — public URL, do not move
LICENSE MIT, for this repository's own code
marketplace.toml plugin marketplace source of truth
mcp-registry.toml curated portable MCP registry
memory/ persistent memory files
migrations/ versioned bootstrap migrations run by `fkt`
package.json catalog toolchain scripts
providers/ API provider definitions
PSScriptAnalyzerSettings.psd1 PowerShell lint configuration
README.md this file
scripts/ maintenance and test scripts
security-advisories.tsv advisory feed consumed by `fkt`
SECURITY.md vulnerability reporting
settings.base.json repo-owned safety config merged on every provider switch
settings.json.example optional settings, merged with settings.base.json on install
skills/ repo-owned plugins (fk-*) and installer-fetched packs
skills-source.lock.json pinned revisions and digests
skills-sources.toml declarative source manifest
tsconfig.json TypeScript configuration
utils/ shared Python helpers
VERSION single version source of truth
Before opening a pull request:
bun install --frozen-lockfile
bun run typecheck
bun test catalog/tests
bun run catalog:check
bun run docs:checkSee CONTRIBUTING.md and the Code of Conduct. Report vulnerabilities privately — see SECURITY.md.
This repository's own code is MIT. Cataloged content keeps its upstream licence.
Built on: gstack · Anthropic's skills · impeccable · marketing skills · taste-skill · Karpathy guidelines · rtk