Agent Environment as Code.

Declare MCP servers, skills, and hooks in gandalf.toml. Review before apply. Roll back safely.

  • Declarative gandalf.toml
  • Non-destructive smart merge
  • Instant safety rollback
Install $ brew install qyinm/tap/gandalf
$ gandalf apply --dry-run
📦 Team Manifest: gandalf.toml (version 1.0)
🎯 Target Agents: claude-code, codex
 
🔍 Review Changes before Apply:
+ mcp postgres-db (staging db connector)
+ skill team-reviewer (PR compliance)
Smart Merge Preview
 
Merge team tools into:
• ~/.claude/settings.json
• ~/.codex/config.toml
 
🛡️ Pre-apply safety backup enabled
1-click rollback with gandalf restore
 
enter apply · esc cancel

01  /  Drift proof

Drift shows up late.

MCP, skills, hooks, permissions, config.

parser high
restore safe
env keys only
tracked setup surface
MCP serversskillshookspermissionsagent config
config.toml semantic diff
baseline [model]provider = "openai"model = "gpt-5" [permissions]default = "ask"
working copy [model]provider = "openai"model = "gpt-5-agent" [permissions]default = "allow"

02  /  Control loop

Review every change.

01

Home

$ gandalf

See drift since the latest Save.

changes-first
02

Console

$ 2

Browse setup objects and agent-native sources.

capability shown
03

Review Changes

$ enter

Confirm only the change you reviewed.

no write before review
Supported scope Conservative by default.
Codex + Claude Code user-global setup only
Console rows capability and reason shown
Marketplace entries agent-native source; reviewed install where supported
Saves + restore safety workflow after review

03  /  Install

Install.

No account. 

$ brew install qyinm/tap/gandalf

Installs to Homebrew's bin directory.

Then run gandalf to open Changes-first Home.

04  /  Ready

Make setup reversible.

Review before apply. Roll back safely.

Install GitHub