---

name: g-skl-review description: Code review for security, quality, performance, reusability; structured report with severity and actions. token_budget: low subsystem_memberships: [BUG_AND_QUALITY]

g-skl-review — thinned shim (prompt-layer)

Judgment served by the bundled prompt layer (one canonical copy in .gald3r_sys/engine). Full original text retained in SKILL.full.md for installs without the engine.

What it does: code/work review.

Preferred — fetch the centralized judgment

gald3r prompt get role.code_reviewer · MCP gald3r_prompt_get id=role.code_reviewer

Manual fallback (engine not provisioned)

Follow SKILL.full.md in this directory, plus any rules.md / reference/ / examples/.

---

What is gald3r?

gald3r is a template you drop into any project to give your AI coding assistant a persistent brain.

Once installed, your AI gains:

• Persistent memory across sessions — tasks, bugs, plans, constraints survive every restart
• 110 skills for common dev workflows (code review, QA, task management, planning, and more)
• 179 commands invoked directly in chat (@g-status, @g-go, @g-task-new, @g-bug-report)
• 37 hooks that fire on IDE events (session start, file save, commit)
• 12 rules that keep the agent disciplined every session
• Works in both Cursor and Claude Code over one shared .gald3r/ brain — plan in one, code in the other
• A bundled file-first engine (Mode-A, new in 2.0) — every system (tasks, bugs, vault, releases, …) is driven by a deterministic Python core via the gald3r CLI or an MCP server, with zero LLM calls. gald3r doctor keeps the install healthy. One prerequisite: uv.

Everything is plain markdown files in your repo. No accounts, no API keys beyond what you already have.

What is new in v2.1.0

Coming in the next release

Safe self-update — CLI and in-app

gald3r version-check        # see if a newer gald3r version is available (offline-safe)
gald3r upgrade              # safe update: auto-backup → migrate → rollback on failure
gald3r upgrade --apply      # confirm and apply (--dry-run is the default)
gald3r init --name "My App" # scaffold a fresh gald3r project in any folder
gald3r setup all            # initialize agent + throne against your shared install home

• gald3r version-check queries the gald3r server and reports your version vs. the latest — degrades gracefully offline.
• gald3r upgrade backs up your .gald3r/ folder with a timestamp, migrates it to the latest format, and rolls back byte-for-byte if anything goes wrong. Your tasks, bugs, and plans are never touched.
• Throne in-app update — gald3r_throne shows an "update available" badge when a new version is detected and can apply the update entirely from within the app (no Python required, compiled Rust updater with full backup + rollback).
• Centralized install home + a global gald3r command, with a USB-portable mode (--portable).
• gald3r init / gald3r update scaffold and update projects in any folder, idempotently.

Install the apps: gald3r install agent and gald3r install throne download the precompiled, signed apps from our public GitHub Releases (Gald3r-Labs/gald3r_agent and Gald3r-Labs/gald3r_throne) and verify them before installing -- the agent binary against a SHA-256 sidecar, the Throne installer against a minisign .sig (a missing/tampered signature fails loud, never a silent install). Use --dry-run to preview, --release vX.Y.Z to pin a version, or --from-source for a local developer build (see RELEASE.md). macOS is coming soon.

---

Quick Install

Option 1 — Copy the template (recommended)

git clone https://github.com/wrm3/gald3r.git

# Default: installs Cursor + Claude Code + shared brain
cp -r gald3r/project_template/. /path/to/your/project/

Then open your project in Cursor or Claude Code and run @g-setup / /g-setup.

Option 2 — Installer script (supports all 34 platforms)

# Windows: double-click or run the .bat (forwards to Python)
.\setup_gald3r_project.bat -TargetPath "C:\MyProject"

# Install for a specific platform only
.\setup_gald3r_project.bat -TargetPath "C:\MyProject" -Platform windsurf
.\setup_gald3r_project.bat -TargetPath "C:\MyProject" -Platform cline
.\setup_gald3r_project.bat -TargetPath "C:\MyProject" -Platform cursor    # Cursor only (no .claude/)

# macOS/Linux (or any platform): run the Python installer directly
python setup_gald3r_project.py --target-path "/path/to/MyProject"

# Install for a specific platform only
python setup_gald3r_project.py --target-path "/path/to/MyProject" --platform windsurf
python setup_gald3r_project.py --target-path "/path/to/MyProject" --platform cline
python setup_gald3r_project.py --target-path "/path/to/MyProject" --platform cursor    # Cursor only (no .claude/)

---

What Gets Installed

Default install (Cursor + Claude Code):

your-project/
├── .cursor/          ← Cursor config (rules, skills, commands, hooks, agents)
├── .claude/          ← Claude Code config (same skill set, markdown format)
├── .gald3r/          ← Shared project memory (tasks, bugs, plans, constraints)
├── .gald3r_sys/      ← gald3r system files (skills engine, platform specs)
├── AGENTS.md         ← Universal agent instructions (read by both IDEs)
├── CLAUDE.md         ← Claude Code entry point
└── WORKFLOW.md       ← Project workflow definition

Platform-specific install (e.g. -Platform windsurf): same shared brain, plus the platform's config folder (.windsurf/rules/ etc.). Cursor and Claude config are skipped.

---

The engine (CLI + MCP)

New in 2.0: a bundled, file-first Python engine drives every system deterministically — no LLM, no network, no Docker. It lives in .gald3r_sys/engine/. The only prerequisite is uv; the first run provisions it automatically.

Command line

# from your project root — the first run builds the engine (a few seconds), then it's instant
uv run --project .gald3r_sys/engine gald3r doctor          # health check  (add --fail-below 80 for CI)

For brevity, alias the prefix — alias gald3r='uv run --project .gald3r_sys/engine gald3r' — then:

gald3r task new   --title "Wire up auth" --priority high
gald3r bug new    --title "Login 500 on empty cart" --severity high
gald3r goal add   --text "Ship the MVP by Friday"
gald3r vault ingest --title "JWT notes" --type article --source https://example.com/jwt
gald3r inbox                                               # absorb staged task/bug drafts into live state
gald3r prompt get role.code_reviewer                       # load a reasoning brief
gald3r --json task list                                    # machine-readable output

(python -m gald3r … works too, if you'd rather not alias.)

As an MCP server

Expose the same operations as ~20 MCP tools to any MCP-capable agent. Add to your client's MCP config:

{
  "mcpServers": {
    "gald3r": {
      "command": "uv",
      "args": ["run", "--project", ".gald3r_sys/engine", "gald3r", "mcp"]
    }
  }
}

Your agent then calls gald3r_task_new, gald3r_bug_list, gald3r_prompt_get, … directly.

No engine? Still works.

The engine is additive. The .gald3r/ state is plain markdown, and every slimmed skill keeps a SKILL.full.md fallback — so a files-only install (no uv / Python) runs unchanged.

---

Platform Support

Platform	Tier	Notes
Cursor	✅ Tier 1	Rules (.mdc), skills, commands, hooks, agents
Claude Code	✅ Tier 1	Rules (.md), skills, commands, hooks, agents
Windsurf, Cline, Roo, Aider	⚠️ Tier 2	Rules + brain + AGENTS.md
Copilot, Codex, Augment, Gemini, Qwen, Continue	⚠️ Tier 2	Rules + brain + AGENTS.md
20+ more	🔜 Tier 3	Brain + AGENTS.md (rules where supported)

Use -Platform <name> with the installer, or copy platforms/<name>/ manually. See gald3r_supported_platforms.html for the full list.

Cursor + Claude Code users get the full experience. Other platforms receive the shared .gald3r/ brain and AGENTS.md instructions, with rules where the platform supports them.

---

How It Works

Your project root
├── AGENTS.md  ─────────────────────────────────────────────────────────┐
├── .cursor/ (rules + skills + commands)   ← Cursor reads these         │
├── .claude/ (rules + skills + commands)   ← Claude Code reads these     │
│                                                                         │
└── .gald3r/ ────────────────────────────────────────────────────────────┘
    TASKS.md    ← shared task list, visible to both IDEs
    BUGS.md     ← shared bug tracker
    PLAN.md     ← shared strategy & milestones
    CONSTRAINTS.md  ← rules the agent must never break

Every command you run in Cursor or Claude Code reads and writes these same files. Switch between tools anytime — context is never lost.

---

Key Commands

Command	What it does
@g-setup / /g-setup	Initialize gald3r in a new project
@g-install-agent / /g-install-agent	Download + install the Gald3r Agent CLI from the public GitHub Release
@g-install-throne / /g-install-throne	Download + install the Gald3r Throne desktop app from the public GitHub Release
@g-status / /g-status	Show project health: tasks, bugs, open items
@g-go / /g-go	Start an autonomous work session on the next task
@g-task-new	Create a new task with spec
@g-bug-report	File and triage a bug
@g-medic	Run self-diagnostics on the gald3r installation
@g-plan	Update and review the project plan

Full command catalog: gald3r Wiki — Commands

---

Project Structure After Install

.gald3r/
├── TASKS.md          ← master task index
├── BUGS.md           ← bug index
├── PLAN.md           ← milestones and strategy
├── PROJECT.md        ← vision, mission, goals
├── CONSTRAINTS.md    ← things the AI must never do
├── SUBSYSTEMS.md     ← component registry
├── tasks/            ← individual task files (one per task)
├── bugs/             ← individual bug files
└── features/         ← PRD files

---

Contributing

See CONTRIBUTING.md. Issues and PRs welcome.

---

License

Fair Source License 1.1 (FSL-1.1-Apache) — see NOTICE for third-party attributions.

---

Powered by gald3r v3.0.0 · Changelog · Roadmap