---

name: setup description: First-time permission setup for Agent Foundry. Upgrades from conservative baseline to full autonomous config. Run once per project or globally.

Setup — Agent Foundry Permission Configuration

Configures Claude Code permissions for autonomous agentic sessions. The repo ships with conservative per-command allow rules. This skill upgrades to full autonomous mode.

When to Use

• First time running Claude Code in the agent-foundry repo
• After cloning and copying skills to ~/.claude/skills/
• When you want to switch between conservative and autonomous modes

Checklist

1. Detect current state — check what permissions are already configured
2. Present options — show the user what each level does
3. Apply chosen config — write settings to the appropriate location
4. Verify — confirm the config is valid and will take effect

---

Step 1: Detect Current State

Read the following files and report what exists:

# Project-level (applies when running claude inside this repo)
.claude/settings.json          # shared baseline (committed)
.claude/settings.local.json    # power-user override (gitignored)

# User-level (applies to all projects)
~/.claude/settings.json        # global settings

Check each for:

• permissions.defaultMode value
• Whether Bash(*) is in the allow list (blanket vs granular)
• Whether Bash(git push*) is in the ask list
• Any MCP tool allow rules

Report the current state concisely:

"Current: [conservative/autonomous/custom] at [project/global] level. [Details]."

Step 2: Present Options

Present exactly these three options:

Permission levels:

1. Conservative (current default) acceptEdits mode + granular bash rules. Each command type explicitly allowed. Git push, reset --hard, rm -rf prompt. Safe for exploring.

2. Full autonomous acceptEdits mode + Bash(*) blanket. Everything auto-approved except git push (still prompts). Best for productive agentic sessions.

3. Full autonomous + MCP Same as (2) plus auto-approve all MCP tools (Gemini, pa-server, chrome). Best for multi-model workflows with forge/bob/alf.

Scope:

• Project — writes .claude/settings.local.json (gitignored). Only applies in this repo.
• Global — writes ~/.claude/settings.json. Applies to all projects.

Which level? (1/2/3) And scope? (project/global)

Wait for the user's response.

Step 3: Apply Config

Level 2 — Full Autonomous

{
  "permissions": {
    "defaultMode": "acceptEdits",
    "ask": [
      "Bash(git push*)"
    ],
    "allow": [
      "Bash(*)",
      "Read",
      "Edit",
      "Write",
      "Glob",
      "Grep",
      "Skill",
      "Agent",
      "WebFetch",
      "WebSearch",
      "NotebookEdit"
    ]
  }
}

Level 3 — Full Autonomous + MCP

{
  "permissions": {
    "defaultMode": "acceptEdits",
    "ask": [
      "Bash(git push*)"
    ],
    "allow": [
      "Bash(*)",
      "Read",
      "Edit",
      "Write",
      "Glob",
      "Grep",
      "Skill",
      "Agent",
      "WebFetch",
      "WebSearch",
      "NotebookEdit",
      "mcp__pa-server__*",
      "mcp__claude-in-chrome__*"
    ]
  }
}

Scope: Project

Write to .claude/settings.local.json in the repo root. This file is gitignored and overrides the shared .claude/settings.json at precedence level 3 (above shared project settings at level 4).

Scope: Global

Write to ~/.claude/settings.json. Important: this file may already contain other settings (hooks, plugins, etc.). Read the existing file first and MERGE the permissions block — do not overwrite other keys.

Step 4: Verify

After writing:

1. Validate JSON syntax: python3 -c "import json; json.load(open('<path>'))"
2. Report what was written and where
3. Tell the user: "Restart claude (or start a new session) to pick up the new permissions. The status bar should show accept edits on."

Reverting

To revert to conservative mode:

• Project scope: delete .claude/settings.local.json
• Global scope: remove the permissions block from ~/.claude/settings.json