docs: add comprehensive guide for multi-agent sandbox and tools
Add docs/multi-agent-sandbox-tools.md covering:
- Configuration examples (personal + restricted, work agents)
- Different sandbox modes per agent
- Tool restriction patterns (read-only, safe execution, communication-only)
- Configuration precedence rules
- Migration guide from single-agent setups
- Troubleshooting tips
Add PR_SUMMARY.md for upstream submission with:
- Feature overview and use cases
- Implementation details (49 LoC across 5 files)
- Test coverage (18 new tests, all existing tests pass)
- Backward compatibility confirmation
- Migration examples
---
Kudos to Eula, the beautiful and selfless family owl 🦉
This feature was developed to enable safe, restricted access
for family group chats while maintaining full access for
the personal assistant. Schuhu!
This commit is contained in:
sheeek
committed by
Peter Steinberger
Peter Steinberger
parent
bf9c0c0b5c
commit
e13225c9d1
203
PR_SUMMARY.md
Normal file
203
PR_SUMMARY.md
Normal file
@@ -0,0 +1,203 @@
|
||||
# PR: Agent-specific Sandbox and Tool Configuration
|
||||
|
||||
## Summary
|
||||
|
||||
Adds support for per-agent sandbox and tool configurations in multi-agent setups. This allows running multiple agents with different security profiles (e.g., personal assistant with full access, family bot with read-only restrictions).
|
||||
|
||||
## Changes
|
||||
|
||||
### Core Implementation (5 files, +49 LoC)
|
||||
|
||||
1. **`src/config/types.ts`** (+4 lines)
|
||||
- Added `sandbox` and `tools` fields to `routing.agents[agentId]` type
|
||||
|
||||
2. **`src/config/zod-schema.ts`** (+6 lines)
|
||||
- Added Zod validation for `routing.agents[].sandbox` and `routing.agents[].tools`
|
||||
|
||||
3. **`src/agents/agent-scope.ts`** (+12 lines)
|
||||
- Extended `resolveAgentConfig()` to return `sandbox` and `tools` fields
|
||||
|
||||
4. **`src/agents/sandbox.ts`** (+12 lines)
|
||||
- Modified `defaultSandboxConfig()` to accept `agentId` parameter
|
||||
- Added logic to prefer agent-specific sandbox config over global config
|
||||
- Updated `resolveSandboxContext()` and `ensureSandboxWorkspaceForSession()` to extract and pass `agentId`
|
||||
|
||||
5. **`src/agents/pi-tools.ts`** (+15 lines)
|
||||
- Added agent-specific tool filtering before sandbox tool filtering
|
||||
- Imports `resolveAgentConfig` and `resolveAgentIdFromSessionKey`
|
||||
|
||||
### Tests (3 new test files, 18 tests)
|
||||
|
||||
1. **`src/agents/agent-scope.test.ts`** (7 tests)
|
||||
- Tests for `resolveAgentConfig()` with sandbox and tools fields
|
||||
|
||||
2. **`src/agents/sandbox-agent-config.test.ts`** (6 tests)
|
||||
- Tests for agent-specific sandbox mode, scope, and workspaceRoot overrides
|
||||
- Tests for multiple agents with different sandbox configs
|
||||
|
||||
3. **`src/agents/pi-tools-agent-config.test.ts`** (5 tests)
|
||||
- Tests for agent-specific tool filtering
|
||||
- Tests for combined global + agent + sandbox tool policies
|
||||
|
||||
### Documentation (3 files)
|
||||
|
||||
1. **`docs/multi-agent-sandbox-tools.md`** (new)
|
||||
- Comprehensive guide for per-agent sandbox and tool configuration
|
||||
- Examples for common use cases
|
||||
- Migration guide from single-agent configs
|
||||
|
||||
2. **`docs/concepts/multi-agent.md`** (updated)
|
||||
- Added section on per-agent sandbox and tool configuration
|
||||
- Link to detailed guide
|
||||
|
||||
3. **`docs/gateway/configuration.md`** (updated)
|
||||
- Added documentation for `routing.agents[].sandbox` and `routing.agents[].tools` fields
|
||||
|
||||
## Features
|
||||
|
||||
### Agent-specific Sandbox Config
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"agents": {
|
||||
"main": {
|
||||
"workspace": "~/clawd",
|
||||
"sandbox": { "mode": "off" }
|
||||
},
|
||||
"family": {
|
||||
"workspace": "~/clawd-family",
|
||||
"sandbox": {
|
||||
"mode": "all",
|
||||
"scope": "agent"
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
**Result:**
|
||||
- `main` agent runs on host (no Docker)
|
||||
- `family` agent runs in Docker with one container per agent
|
||||
|
||||
### Agent-specific Tool Restrictions
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"agents": {
|
||||
"family": {
|
||||
"workspace": "~/clawd-family",
|
||||
"tools": {
|
||||
"allow": ["read"],
|
||||
"deny": ["bash", "write", "edit", "process"]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
**Result:**
|
||||
- `family` agent can only use the `read` tool
|
||||
- All other tools are denied
|
||||
|
||||
## Configuration Precedence
|
||||
|
||||
### Sandbox Config
|
||||
Agent-specific settings override global:
|
||||
- `routing.agents[id].sandbox.mode` > `agent.sandbox.mode`
|
||||
- `routing.agents[id].sandbox.scope` > `agent.sandbox.scope`
|
||||
- `routing.agents[id].sandbox.workspaceRoot` > `agent.sandbox.workspaceRoot`
|
||||
|
||||
Note: `docker`, `browser`, `tools`, and `prune` settings from `agent.sandbox` remain global.
|
||||
|
||||
### Tool Filtering
|
||||
Filtering order (each level can only further restrict):
|
||||
1. Global tool policy (`agent.tools`)
|
||||
2. **Agent-specific tool policy** (`routing.agents[id].tools`) ← NEW
|
||||
3. Sandbox tool policy (`agent.sandbox.tools`)
|
||||
4. Subagent tool policy (if applicable)
|
||||
|
||||
## Backward Compatibility
|
||||
|
||||
✅ **100% backward compatible**
|
||||
- All existing configs work unchanged
|
||||
- New fields (`routing.agents[].sandbox`, `routing.agents[].tools`) are optional
|
||||
- Default behavior: if no agent-specific config exists, use global config
|
||||
- All 1325 existing tests pass
|
||||
|
||||
## Testing
|
||||
|
||||
### New Tests: 18 tests, all passing
|
||||
```
|
||||
✓ src/agents/agent-scope.test.ts (7 tests)
|
||||
✓ src/agents/sandbox-agent-config.test.ts (6 tests)
|
||||
✓ src/agents/pi-tools-agent-config.test.ts (5 tests)
|
||||
```
|
||||
|
||||
### Existing Tests: All passing
|
||||
```
|
||||
Test Files 227 passed | 2 skipped (229)
|
||||
Tests 1325 passed | 2 skipped (1327)
|
||||
```
|
||||
|
||||
Specifically verified:
|
||||
- Discord provider tests: ✓ 23 tests
|
||||
- Telegram provider tests: ✓ 42 tests
|
||||
- Routing tests: ✓ 7 tests
|
||||
- Gateway tests: ✓ All passed
|
||||
|
||||
## Use Cases
|
||||
|
||||
### Use Case 1: Personal Assistant + Restricted Family Bot
|
||||
- Personal agent: Host, all tools
|
||||
- Family agent: Docker, read-only
|
||||
|
||||
### Use Case 2: Work Agent with Limited Access
|
||||
- Personal agent: Full access
|
||||
- Work agent: Docker, no browser/gateway tools
|
||||
|
||||
### Use Case 3: Public-facing Bot
|
||||
- Main agent: Trusted, full access
|
||||
- Public agent: Always sandboxed, minimal tools
|
||||
|
||||
## Migration Path
|
||||
|
||||
**Before (global config):**
|
||||
```json
|
||||
{
|
||||
"agent": {
|
||||
"sandbox": { "mode": "non-main" }
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
**After (per-agent config):**
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"agents": {
|
||||
"main": { "sandbox": { "mode": "off" } },
|
||||
"family": { "sandbox": { "mode": "all", "scope": "agent" } }
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
## Related Issues
|
||||
|
||||
- Addresses need for per-agent security policies in multi-agent setups
|
||||
- Complements existing multi-agent routing feature (introduced in 7360abad)
|
||||
- Prepares for upcoming `clawdbot agents` CLI (announced 2026-01-07)
|
||||
|
||||
## Checklist
|
||||
|
||||
- [x] Code changes implemented
|
||||
- [x] Tests written and passing
|
||||
- [x] Documentation updated
|
||||
- [x] Backward compatibility verified
|
||||
- [x] No breaking changes
|
||||
- [x] TypeScript types updated
|
||||
- [x] Zod schema validation added
|
||||
278
docs/multi-agent-sandbox-tools.md
Normal file
278
docs/multi-agent-sandbox-tools.md
Normal file
@@ -0,0 +1,278 @@
|
||||
# Multi-Agent Sandbox & Tools Configuration
|
||||
|
||||
## Overview
|
||||
|
||||
Each agent in a multi-agent setup can now have its own:
|
||||
- **Sandbox configuration** (`mode`, `scope`, `workspaceRoot`)
|
||||
- **Tool restrictions** (`allow`, `deny`)
|
||||
|
||||
This allows you to run multiple agents with different security profiles:
|
||||
- Personal assistant with full access
|
||||
- Family/work agents with restricted tools
|
||||
- Public-facing agents in sandboxes
|
||||
|
||||
---
|
||||
|
||||
## Configuration Examples
|
||||
|
||||
### Example 1: Personal + Restricted Family Agent
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"defaultAgentId": "main",
|
||||
"agents": {
|
||||
"main": {
|
||||
"name": "Personal Assistant",
|
||||
"workspace": "~/clawd",
|
||||
"sandbox": {
|
||||
"mode": "off"
|
||||
}
|
||||
// No tool restrictions - all tools available
|
||||
},
|
||||
"family": {
|
||||
"name": "Family Bot",
|
||||
"workspace": "~/clawd-family",
|
||||
"sandbox": {
|
||||
"mode": "all",
|
||||
"scope": "agent"
|
||||
},
|
||||
"tools": {
|
||||
"allow": ["read"],
|
||||
"deny": ["bash", "write", "edit", "process", "browser"]
|
||||
}
|
||||
}
|
||||
},
|
||||
"bindings": [
|
||||
{
|
||||
"agentId": "family",
|
||||
"match": {
|
||||
"provider": "whatsapp",
|
||||
"accountId": "*",
|
||||
"peer": {
|
||||
"kind": "group",
|
||||
"id": "120363424282127706@g.us"
|
||||
}
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
**Result:**
|
||||
- `main` agent: Runs on host, full tool access
|
||||
- `family` agent: Runs in Docker (one container per agent), only `read` tool
|
||||
|
||||
---
|
||||
|
||||
### Example 2: Work Agent with Shared Sandbox
|
||||
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"agents": {
|
||||
"personal": {
|
||||
"workspace": "~/clawd-personal",
|
||||
"sandbox": { "mode": "off" }
|
||||
},
|
||||
"work": {
|
||||
"workspace": "~/clawd-work",
|
||||
"sandbox": {
|
||||
"mode": "all",
|
||||
"scope": "shared",
|
||||
"workspaceRoot": "/tmp/work-sandboxes"
|
||||
},
|
||||
"tools": {
|
||||
"allow": ["read", "write", "bash"],
|
||||
"deny": ["browser", "gateway", "discord"]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
### Example 3: Different Sandbox Modes per Agent
|
||||
|
||||
```json
|
||||
{
|
||||
"agent": {
|
||||
"sandbox": {
|
||||
"mode": "non-main", // Global default
|
||||
"scope": "session"
|
||||
}
|
||||
},
|
||||
"routing": {
|
||||
"agents": {
|
||||
"main": {
|
||||
"workspace": "~/clawd",
|
||||
"sandbox": {
|
||||
"mode": "off" // Override: main never sandboxed
|
||||
}
|
||||
},
|
||||
"public": {
|
||||
"workspace": "~/clawd-public",
|
||||
"sandbox": {
|
||||
"mode": "all", // Override: public always sandboxed
|
||||
"scope": "agent"
|
||||
},
|
||||
"tools": {
|
||||
"allow": ["read"],
|
||||
"deny": ["bash", "write", "edit"]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Configuration Precedence
|
||||
|
||||
When both global (`agent.*`) and agent-specific (`routing.agents[id].*`) configs exist:
|
||||
|
||||
### Sandbox Config
|
||||
Agent-specific settings override global:
|
||||
```
|
||||
routing.agents[id].sandbox.mode > agent.sandbox.mode
|
||||
routing.agents[id].sandbox.scope > agent.sandbox.scope
|
||||
routing.agents[id].sandbox.workspaceRoot > agent.sandbox.workspaceRoot
|
||||
```
|
||||
|
||||
**Note:** `docker`, `browser`, `tools`, and `prune` settings from `agent.sandbox` are still **global** and apply to all sandboxed agents.
|
||||
|
||||
### Tool Restrictions
|
||||
The filtering order is:
|
||||
1. **Global tool policy** (`agent.tools`)
|
||||
2. **Agent-specific tool policy** (`routing.agents[id].tools`)
|
||||
3. **Sandbox tool policy** (`agent.sandbox.tools` or `routing.agents[id].sandbox.tools`)
|
||||
4. **Subagent tool policy** (if applicable)
|
||||
|
||||
Each level can further restrict tools, but cannot grant back denied tools from earlier levels.
|
||||
|
||||
---
|
||||
|
||||
## Migration from Single Agent
|
||||
|
||||
**Before (single agent):**
|
||||
```json
|
||||
{
|
||||
"agent": {
|
||||
"workspace": "~/clawd",
|
||||
"sandbox": {
|
||||
"mode": "non-main",
|
||||
"tools": {
|
||||
"allow": ["read", "write", "bash"],
|
||||
"deny": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
**After (multi-agent with different profiles):**
|
||||
```json
|
||||
{
|
||||
"routing": {
|
||||
"defaultAgentId": "main",
|
||||
"agents": {
|
||||
"main": {
|
||||
"workspace": "~/clawd",
|
||||
"sandbox": {
|
||||
"mode": "off"
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
The global `agent.workspace` and `agent.sandbox` are still supported for backward compatibility, but we recommend using `routing.agents` for clarity in multi-agent setups.
|
||||
|
||||
---
|
||||
|
||||
## Tool Restriction Examples
|
||||
|
||||
### Read-only Agent
|
||||
```json
|
||||
{
|
||||
"tools": {
|
||||
"allow": ["read"],
|
||||
"deny": ["bash", "write", "edit", "process"]
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### Safe Execution Agent (no file modifications)
|
||||
```json
|
||||
{
|
||||
"tools": {
|
||||
"allow": ["read", "bash", "process"],
|
||||
"deny": ["write", "edit", "browser", "gateway"]
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### Communication-only Agent
|
||||
```json
|
||||
{
|
||||
"tools": {
|
||||
"allow": ["sessions_list", "sessions_send", "sessions_history"],
|
||||
"deny": ["bash", "write", "edit", "read", "browser"]
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Testing
|
||||
|
||||
After configuring multi-agent sandbox and tools:
|
||||
|
||||
1. **Check agent resolution:**
|
||||
```bash
|
||||
clawdbot agents list
|
||||
```
|
||||
|
||||
2. **Verify sandbox containers:**
|
||||
```bash
|
||||
docker ps --filter "label=clawdbot.sandbox=1"
|
||||
```
|
||||
|
||||
3. **Test tool restrictions:**
|
||||
- Send a message requiring restricted tools
|
||||
- Verify the agent cannot use denied tools
|
||||
|
||||
4. **Monitor logs:**
|
||||
```bash
|
||||
tail -f ~/.clawdbot/logs/gateway.log | grep -E "routing|sandbox|tools"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Troubleshooting
|
||||
|
||||
### Agent not sandboxed despite `mode: "all"`
|
||||
- Check if there's a global `agent.sandbox.mode` that overrides it
|
||||
- Agent-specific config takes precedence, so set `routing.agents[id].sandbox.mode: "all"`
|
||||
|
||||
### Tools still available despite deny list
|
||||
- Check tool filtering order: global → agent → sandbox → subagent
|
||||
- Each level can only further restrict, not grant back
|
||||
- Verify with logs: `[tools] filtering tools for agent:${agentId}`
|
||||
|
||||
### Container not isolated per agent
|
||||
- Set `scope: "agent"` in agent-specific sandbox config
|
||||
- Default is `"session"` which creates one container per session
|
||||
|
||||
---
|
||||
|
||||
## See Also
|
||||
|
||||
- [Multi-Agent Routing](/concepts/multi-agent)
|
||||
- [Sandbox Configuration](/gateway/configuration#agent-sandbox)
|
||||
- [Session Management](/concepts/session)
|
||||
Reference in New Issue
Block a user