Set up a complete, role-matched AI workflow across Claude Code, Claude Desktop, Google Antigravity, Zed, and OpenCode in 60 seconds—without bloating your context window or overwriting your existing rules.
skill-installer is a cross-platform (macOS, Linux, Windows) interactive terminal wizard and CLI that installs curated tracks of AI skills, /slash commands, adversarial review subagents, and harness rules—preserving your custom CLAUDE.md, GEMINI.md, or AGENTS.md and saving a timestamped rollback snapshot before touching a single file.
Interactive 3-step TUI wizard & command explorer (watch MP4 video)
Dumping 50+ skills into your AI agent all at once backfires: every loaded skill competes for your model's context window and attention, and manual repo scripts often clobber your carefully tuned CLAUDE.md or GEMINI.md.
Without skill-installer |
With skill-installer |
|---|---|
| Context bloat: Loading 50+ skills at once crowds out code context and degrades instruction-following | Role-matched tracks: Pick a curated track (8 to 36 skills) built for how you actually work |
Fragile manual setup: Copying files between ~/.claude/, ~/.gemini/, .agents/, and .opencode/ by hand |
One 60-second wizard: Installs the right layout for Claude Code, Claude Desktop, Antigravity, Zed, or OpenCode |
Clobbered config files: Template scripts overwrite your personal or repo CLAUDE.md / GEMINI.md |
Non-destructive merge + 1-command rollback: Updates a delimited section inside your existing config after saving a backup |
| Serial 1-file subagent loops & cross-harness rule leaks: Reviewers re-read full files one per turn or inherit rules built for a different runtime | Harness-tailored batching & read-only hardening: Pre-packed byte-bounded diff/slice bundles (build_subagent_context.py), parallel read/edit batching, explicit read-only reviewer MCP allowlists, and Antigravity-only sleeping Hydrogen ops routing |
| "Wait, what did that install?": Digging through Markdown files to remember command names | Built-in interactive guide (skill-installer guide): Browse every skill, /slash command, and copy-ready starter prompt in your terminal |
Download the prebuilt binary for your OS from the Latest GitHub Release and run it with no flags to launch the interactive wizard:
# Download the archive for your OS/CPU from Releases, extract, and launch:
tar -xzf skill-installer_<version>_<os>_<arch>.tar.gz
./skill-installermacOS Gatekeeper note: If macOS blocks an unsigned binary downloaded in a browser, clear the quarantine attribute once before running:
xattr -d com.apple.quarantine ./skill-installer
Expand-Archive skill-installer_<version>_windows_amd64.zip -DestinationPath .
.\skill-installer.exegit clone https://github.com/futuregerald/futuregerald-claude-plugin.git
cd futuregerald-claude-plugin
go build -o skill-installer .
./skill-installerEvery loaded skill competes for your AI agent's context window. Instead of installing everything, pick the track designed for your day-to-day work (or cherry-pick individual skills):
| Track | Best For | Skills Installed | Key /slash Commands & Capabilities |
Starter Prompt to Try Immediately |
|---|---|---|---|---|
Opinionated Dev Workflow (--track=opinionated) |
Software engineers shipping production code with high rigor | 32 (Claude / Zed / OpenCode) 36 (Antigravity) |
/start, /start-small (Antigravity), /begin, /lifecycle, /shape, /ui-design, /plan-review, /fix-check, /copy-edit — full 12-phase lifecycle with 3-direction HTML UI prototyping, caller tracing, byte-bounded subagent context bundling (build_subagent_context.py), parallel read/edit batching, TDD, explicit read-only code & security reviewers, and blast-radius verification |
"/start Add passwordless magic-link sign-in to the auth service" |
Just the Basics (--track=basics) |
Developers who want everyday tools without strict phase gates or hooks | 8 | /brainstorming, /future-writing-plans, /systematic-debugging, /test-driven-development, /comprehensive-code-review, /pull-request-description, /handoff, /plugin-update |
"/brainstorming Let's design a webhook retry queue with exponential backoff" |
Product & Strategy Flow (--track=product) |
PMs, designers, founders, and engineering managers | 22 | /shape, /craftsmans-wisdom, /future-grilling, /future-wayfinder, /grilling-page, /write-a-prd, /writing-tickets, /ticket-grooming, /ui-design, /visualize-plan, /copy-edit, /editor-pass, /meeting-debrief, /team-pulse |
"/shape We want to redesign customer onboarding so teams activate in under 5 minutes" |
Copy, Brand & UI Polish (--track=copy-polish) |
Auditing and elevating landing pages, product UI, copy, and motion | 9 | /copy-edit, /copywriting, /marketing-psychology, /editor-pass, /impeccable, /no-slop-ui, /huashu-design, /animation-principles, /plugin-update |
"/copy-edit Audit our pricing page for clarity, conversion friction, and visual hierarchy" |
Pick Individual Skills (--track=custom) |
Power users building a custom à la carte toolkit | Any subset of 51 (Claude / Zed / OpenCode) or 55 (Antigravity) |
Interactive skill browser with [x] multi-select, category badges, and plain-English descriptions |
Any starter prompt from your chosen skills |
Running ./skill-installer with no arguments opens the SkillBot terminal UI:
Choose a 60-second guided setup, open the interactive command guide, pull updates from main, restore a timestamped backup, or cleanly uninstall.
Select Claude Code (CLI & Claude Desktop), Google Antigravity (Desktop IDE), Zed Editor, or OpenCode, then choose whether to install Globally (~) or into a Specific Project Folder (e.g., ~/code/acme-api) so you can trial a track in one repo without touching your global setup.
Compare the 5 tracks with plain-English summaries, skill counts, and a recommendation badge. Press p on any track to inspect every included skill, or press d / Enter on a skill to expand its full description and copy-ready starter prompt before installing.
If skill-installer detects an existing CLAUDE.md, GEMINI.md, or AGENTS.md, it never blindly overwrites your file. You choose how to handle it:
- Add / update workflow section in existing config + save backup (Recommended) — Keeps every line of your existing rules intact and appends/updates a managed block (
<!-- BEGIN SKILL-INSTALLER WORKFLOW RULES -->…<!-- END SKILL-INSTALLER WORKFLOW RULES -->) after saving a full backup snapshot. - Replace config with clean track template + save backup — Snapshots your existing file first, then writes the clean track template.
- Keep my existing config untouched — Installs skills, subagents, and
/slashcommands without modifying yourCLAUDE.md/GEMINI.md/AGENTS.md. - Replace without saving a backup — Overwrites directly.
Every run finishes with exact file paths, your rollback snapshot ID, and copy-ready starter prompts tailored to your installed track so you can paste one directly into your next AI session.
Run skill-installer guide at any time (or open it directly from the Setup Complete screen) for an interactive terminal tour of your installed toolkit:
- Tab 1 (
1. What You Can Type Now): Plain-English breakdown of/slashcommands vs. auto-loaded skills for your specific harness, plus copy-ready prompts. - Tab 2 (
2. How This Toolkit Works): Explains why curated tracks protect your context window, how skills vs. subagents vs. rules fit together, and how rollback backups work. - Tab 3 (
3. Skill & Command Explorer): Searchable browser of your track's skills (pressato toggle between your installed track and all 55 catalog skills) with category badges,/slashtriggers, full descriptions, and starter prompts.
# Open the interactive 3-tab guide
skill-installer guide
# Print a plain-text summary for scripts or non-TTY terminals
skill-installer guide --target=claude --track=opinionated --yesskill-installer adapts every skill, subagent, /slash command, and rule file to each harness's native directory structure—and coexists cleanly with your existing desktop app settings and MCP servers:
| Harness | CLI & Desktop Coverage | Global Install Paths | Project Install Paths | Coexistence Guarantees |
|---|---|---|---|---|
Claude Code & Claude Desktop (--target=claude) |
claude CLI + Claude Desktop (Claude.app Code / Cowork tab) |
~/.claude/CLAUDE.md~/.claude/skills/~/.claude/agents/~/.claude/commands/ |
<project>/CLAUDE.md<project>/.claude/skills/<project>/.claude/agents/<project>/.claude/commands/ |
Preserves claude_desktop_config.json (MCP servers & desktop preferences) untouched; installs explicit read-only reviewer MCP allowlists and effort: low search/test helpers |
Google Antigravity (--target=antigravity) |
Antigravity Desktop IDE (Antigravity.app) + agy CLI |
~/.gemini/GEMINI.md~/.gemini/config/skills/~/.gemini/config/plugins/antigravity-subagents/~/.gemini/config/hooks/ & ~/.gemini/config/hooks.json (Opinionated track)~/.gemini/config/rules/org.example.md |
<project>/GEMINI.md<project>/.agents/skills/<project>/.agents/agents/ |
Preserves ~/.gemini/antigravity/mcp_config.json, merges co-tenant PreToolUse entries in ~/.gemini/config/hooks.json, and includes Carbon Medium / /start-small (Argon) + sleeping Hydrogen (0.2x) ops routing |
Zed Editor (--target=zed) |
Zed Desktop (Zed.app / zed CLI) |
~/.config/zed/AGENTS.md (macOS/Linux)%APPDATA%\Zed\AGENTS.md (Windows)~/.agents/skills/ |
<project>/AGENTS.md<project>/.agents/skills/ |
Preserves ~/.config/zed/settings.json and keymap.json; skills surface automatically as native /<skill-name> slash commands |
OpenCode (--target=opencode) |
opencode CLI & TUI |
~/.config/opencode/AGENTS.md~/.config/opencode/skills/~/.config/opencode/agents/~/.config/opencode/commands/ |
<project>/AGENTS.md<project>/.opencode/skills/<project>/.opencode/agents/<project>/.opencode/commands/ |
Preserves ~/.config/opencode/opencode.json model and provider config untouched; installs validated YAML boolean tools: maps on read-only reviewers |
Every action in the interactive wizard is also available via flags for dotfiles, onboarding scripts, and headless environments:
# Install the Opinionated Dev Workflow globally for Claude Code + Claude Desktop
skill-installer --target=claude --global --track=opinionated --backup=true --yes
# Non-destructively merge the workflow block into an existing CLAUDE.md / GEMINI.md / AGENTS.md
skill-installer --target=claude --global --track=opinionated --merge-config --backup=true --yes
# Install the Product & Strategy track into a single project repository
skill-installer --target=claude --project=/path/to/repo --track=product --yes
# Install a custom subset of skills for Zed Editor
skill-installer --target=zed --global --track=custom \
--skills=brainstorming,future-writing-plans,test-driven-development,comprehensive-code-review --yes
# List all tracks and catalog skills (or filter by --target / --track)
skill-installer list --target=claude
# Check for and apply upstream skill updates from main (saves a backup first)
skill-installer update --apply --yes
# Restore the latest safety backup (or pass --id=<backup-id>)
skill-installer restore --yes
# Uninstall plugin-managed files while keeping your custom skills untouched
skill-installer uninstall --target=claude --global --yesSkill, /slash command, subagent, and workflow rule updates ship continuously on main—you do not need to wait for a new GitHub Release or re-download the binary to get new or updated skills:
- From your AI chat session: Type
/plugin-updatein Claude Code, Antigravity, Zed, or OpenCode. - From your terminal: Run
skill-installer update(or choose Check for skill & workflow updates on the home screen). It compares your installed commit SHA against the latest commit onmain, lists the exact commits and skills that changed, saves a rollback snapshot, and refreshes your installed track directly frommain. - Upgrading the CLI binary itself (major releases only): GitHub Releases are only cut when the
skill-installerbinary itself has major changes. When a new binary release is available,skill-installer updatewill let you know and you can runskill-installer update --selfto upgrade the executable in place.
Every backup snapshot is stored under ~/.config/skill-installer/backups/<timestamp>-<seq>-<harness>-<scope>/ (or %APPDATA%\skill-installer\backups\ on Windows) with a complete manifest of pre-existing and newly created files. Running skill-installer restore restores modified files and removes newly added ones in one step.
Each installation writes a receipt tracking only the files managed by skill-installer. Running skill-installer uninstall removes those managed files (or strips only the <!-- BEGIN SKILL-INSTALLER WORKFLOW RULES --> block if you merged into an existing CLAUDE.md / GEMINI.md / AGENTS.md), leaving your own custom skills and rules untouched.
test -z "$(gofmt -l .)"
go vet ./...
HOME=$(mktemp -d) go test -race ./...
PYTHONDONTWRITEBYTECODE=1 pytest -p no:cacheprovider -q \
internal/catalog/shared/skills/editor-pass/scripts \
internal/catalog/shared/skills/future-skill-creator/scripts \
internal/catalog/shared/skills/grilling-page/scripts \
internal/catalog/shared/skills/subagent-driven-development/scripts \
internal/catalog/shared/skills/team-pulse/scripts \
internal/catalog/shared/skills/visualize-plan/scripts \
internal/catalog/antigravity/hooksTo verify skill-installer against real installed CLI & desktop harnesses (claude, Claude.app, Antigravity.app, Zed.app, opencode), live PreToolUse circuit-breaker hook execution, and real pseudo-terminal (PTY) Bubble Tea keystroke journeys inside an isolated Linux container:
make test-e2e





