Skip to content

trajectoryRL/openclaw

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16,857 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

OpenClaw (TrajectoryRL Fork)

Fork of OpenClaw configured as a mock evaluation environment for the TrajectoryRL project.

This fork is not intended for general use. It provides a sandboxed OpenClaw instance with a comprehensive mock tool library for reproducible A/B testing of AGENTS.md policies.

Changes from upstream

1. extensions/clawbench-tools/

An OpenClaw plugin that registers 25 mock tools across 8 categories:

Category Tools Count
Email & Inbox inbox_list, email_read, email_draft, email_send, email_archive 5
Calendar calendar_read, calendar_create, calendar_update, calendar_delete 4
Messaging (Slack) slack_list_channels, slack_read_messages, slack_post_message, slack_send_dm 4
Tasks (Jira/Linear) task_list, task_get, task_create, task_update 4
Documents (Drive/Notion) doc_list, doc_read, doc_create 3
Contacts contacts_list, contacts_get 2
Memory / Notes memory_read, memory_write 2
Web Search search_web 1

Each tool proxies HTTP requests to an external mock server that returns deterministic fixture data. The plugin always registers all tools — scenarios control which subset is active via the tools.allow config.

2. sandbox-config/openclaw.json

Pre-configured settings that skip onboarding and lock down the tool environment:

  • Gateway: LAN bind, token auth, OpenAI-compatible HTTP API enabled
  • Tools: all built-in tools denied (exec, browser, web_search, etc.)
  • All 25 mock tools + read + session_status allowed (base config)
  • Plugin enabled and pointed at the mock server

The model is controlled by the CLAWBENCH_DEFAULT_MODEL env var (default: anthropic/claude-sonnet-4-6). The config uses ${CLAWBENCH_DEFAULT_MODEL} as a placeholder, resolved at container startup. See ClawBench docs for local LLM setup.

Note: In practice, the clawbench harness generates a scenario-specific openclaw.json at runtime that only allows the tools needed for that scenario. The base config here is a fallback that allows everything.

3. Dockerfile.clawbench

Custom Docker image that:

  • Builds OpenClaw from source
  • Bakes in sandbox-config/openclaw.json (no interactive setup)
  • Starts the gateway with --allow-unconfigured --bind lan

Usage

This fork is used by the clawbench harness. See that repo for setup and usage instructions.

Upstream

For the original OpenClaw documentation, see github.com/openclaw/openclaw.

License

Same as upstream: MIT

About

Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞

Resources

License

Contributing

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • TypeScript 86.5%
  • Swift 8.9%
  • Kotlin 2.1%
  • Shell 1.0%
  • JavaScript 0.7%
  • CSS 0.4%
  • Other 0.4%