Skip to content

OpenClaw for ConnectSoft (Overview)

This document explains how OpenClaw fits into ConnectSoft workflows and what boundaries we enforce to keep it safe, predictable, and cost-controlled.

When to use OpenClaw vs Cursor vs Factory agents

Use the right tool for the job:

  • OpenClaw: long-running, always-on personal assistant with channels (chat UI, Telegram, etc.), capable of orchestrating repeatable workflows (repo sync, indexing, scaffolding, builds) on a remote workstation.
  • Cursor: interactive implementation and refactoring in a repo with tight review loops and precise diffs.
  • Factory agents: standardized, repeatable generation workflows across many artifacts (services, libraries, docs) when you want “platform-level consistency”.

Tip

A common pattern is OpenClaw to orchestrate, Cursor to implement, and Factory agents to standardize/scale.

Safety boundaries (non-negotiable)

Secrets and sensitive data

  • Do not paste PATs, API keys, certificates, customer data, or production secrets into OpenClaw chats, configs, prompts, or logs.
  • Use environment variables and secret stores on the remote machine.

Warning

Treat agent transcripts and run logs as potentially shareable. If it shouldn’t be in a PR, it shouldn’t be in the agent.

Approval gates

OpenClaw must require explicit approval before:

  • pushing to any remote git repo
  • publishing packages (NuGet, containers, artifacts)
  • changing shared infrastructure or pipelines
  • editing non-temp “production” workspaces

Repo and filesystem boundaries

OpenClaw work must be constrained to:

  • a repo allowlist (ConnectSoft repos you explicitly permit)
  • a workspace root (e.g., C:\\OpenClaw\\workspaces\\connectsoft\\)
  • a runs/temp folder for generated outputs (never generate into existing repos unless asked)

High-level flow

flowchart TB
  user[You] --> channel[Channel_UI]
  channel --> gateway[OpenClaw_Gateway]
  gateway --> assistants[ConnectSoft_Assistants]
  assistants --> repoSync[Repo_Sync_ADO]
  assistants --> indexer[Semantic_Index]
  assistants --> tempRuns[Temp_Runs_Folder]
  assistants --> report[Run_Summary]
Hold "Alt" / "Option" to enable pan & zoom