Small-Step Collaboration
Make an agent guide users one smallest useful action at a time instead of dumping long checklists. Best for webmaster tools, deployments, console debugging, and onboarding flows.
Copy the extracted small-step-collaboration folder into ~/.codex/skills, ~/.openclaw/skills, or ~/.hermes/skills What this skill fixes
Many agents are not blocked by capability. They are blocked by communication overload.
Users usually need the next click, the next command, or the next thing to check. They do not need a giant wall of instructions that becomes stale as soon as the screen changes.
This skill moves the agent into a tighter collaboration loop:
- give one smallest useful action
- ask for one result artifact
- interpret the result
- continue with the next step
Best-fit use cases
- Google Search Console, Bing Webmaster Tools, and Baidu Search Resource Platform
- Cloud consoles such as Cloudflare, Vercel, GitHub, and database dashboards
- setup, deployment, indexing, verification, and troubleshooting flows
- any task where the human must click, inspect, or confirm something the agent cannot do alone
Core behavior
With this skill enabled, the agent should:
- default to one meaningful next step per turn
- use exact button, field, menu, and command names
- ask for only one screenshot, output block, or yes/no result at a time
- explain what the current check is proving
- pause at the smallest safe checkpoint when risk is involved
Supported runtimes
This bundle is structured around a shared SKILL.md layout and is currently aimed at:
- Codex
- OpenClaw
- Hermes
Claude Code can reuse the same collaboration rules later with a platform-specific wrapper.
Install
Codex
Copy the extracted small-step-collaboration folder into:
~/.codex/skills/
OpenClaw
Copy the same folder into:
~/.openclaw/skills/
Hermes
Copy the same folder into:
~/.hermes/skills/
Bundle contents
SKILL.mdfor trigger and behavior rulesagents/openai.yamlfor Codex UI metadatareferences/platform-notes.mdfor platform install notesreferences/response-patterns.mdfor reusable response patterns
This is a good fit as a communication-layer skill that sits on top of your debugging, deployment, or console-navigation workflows.