Product Principles & Best Practices
Clawforce is designed for easy-to-use agent team setup and background automation. This guide outlines product philosophy and recommended practices.
Our Focus: Easy Agent Team Setup
Clawforce prioritizes:
- 1-Click Deployment — Deploy pre-built agents from the marketplace; no code required
- Visual Configuration — Set objectives, tools, and integrations via the dashboard
- Team Coordination — Plans, A2A communication, and shared workspaces
- Low Friction — From idea to running agent in minutes
Background Agents
Clawforce agents are background workers — they run 24/7 in isolated containers, triggered by cron, events, or channels. We want to keep it simple and run purely in the background. Therefore. We do not introduce:
- Computer Use — Direct mouse/keyboard control of desktops
- Browser Use — Automated browser driving (including headless mode)
Best Practices
1. Security First
- Use container isolation — Docker/Podman per agent; avoid process runtime in production
- Sandboxed mode when possible — Read-only rootfs, no network for sensitive workloads
- Secrets — Use the vault; never hardcode API keys
- Approval gates — Enable
ask_before_runfor sensitive tools (exec,write_file, etc.) - Audit logs — Capture and monitor
clawforce.auditfor security events
See Security for details.
2. Trusted Software & Registries
- Software Registry — Use trusted, curated software from the marketplace
- MCP Registry — Prefer official or internal MCP servers over arbitrary HTTP endpoints
- Avoid unvetted tools — Limit agents to software and MCPs you’ve reviewed
3. Use Internal MaaS (MCP-as-a-Service)
- Internal MaaS — Point agents to your internal MCP servers for governance, cost control, and compliance
- LiteLLM — Works with any OpenAI-compatible endpoint
- Data residency — Keeps prompts and responses within your infrastructure
4. Skills Registry
- Reuse skills — Use the Skills Registry to share capabilities across agents
- Version control — Treat skills as code; version and review changes
- Custom skills — Add domain-specific skills rather than overloading prompts
5. Plans for Team Coordination
- Multi-agent work — Use Plans (Kanban-style boards) when several agents collaborate
- Clear boundaries — Define roles and responsibilities per agent
- External sync — Pull tasks from GitHub, Jira, etc.; Plans orchestrate execution
6. Start Strict, Relax as Needed
- Start sandboxed — Test new agents in Safe mode first
- Promote gradually — Move to Standard (permissive) only when network/software is required
- Privileged — Reserve for trusted, controlled environments
7. Review & Iterate
- Activity logs — Regularly inspect agent behavior
- Workspace artifacts — Review outputs before trusting automation
- Iterate on prompts — Refine objectives based on real runs
Summary
| Principle | Action |
|---|---|
| Easy setup | Deploy from marketplace, configure via dashboard |
| Background-first | No Computer/Browser Use; use MCP for integrations |
| Security | Containers, vault, approval gates, audit logs |
| Trusted tooling | Curated software, MCP registry, internal MaaS (MCP-as-a-Service) |
| Skills & Plans | Skills Registry for reuse; Plans for team coordination |
| Gradual trust | Start sandboxed; relax only when necessary |
For more, see Configuration, Security, and Plans.