Skip to content

openclaw.md

Latest commit

 

History

History
124 lines (94 loc) · 3.43 KB

openclaw.md

File metadata and controls

124 lines (94 loc) · 3.43 KB

Install For OpenClaw

OpenClaw uses an agent workspace with workspace files and skill folders.

Official references:

Workspace Instructions

Put the Agents Remember instruction in the OpenClaw workspace AGENTS.md, pointing at the actual coordination runtime path OpenClaw can read:

# Workspace Agent Instructions

Read and follow `/path/to/ar-coordination/AGENTS.md` before working in any target project.
Treat these rules as workspace instructions!

@/path/to/ar-coordination/AGENTS.md

OpenClaw workspaces may contain other standing instruction files. Keep Agents Remember focused on repository memory and task workflow rules; do not put secrets in workspace docs.

Skills

Install the runtime:

python3 agents-remember-md/installer/install-runtime.py /path/to/ar-coordination

OpenClaw commonly uses global skills under ~/.openclaw/skills/ and workspace skills under <workspace>/skills/, with workspace skills taking precedence.

Agents Remember skills should stay linked to the installed runtime under ar-coordination/skills. Use the default nested/tree install layout. Do not work around OpenClaw skill discovery by copying the skills into the OpenClaw root or by using flat layout: several bundled scripts use their runtime layout to find shared modules and default coordination paths. A copied or flattened tree can make the skills visible while making commands such as C-08-ar-coordination-context-resolver resolve the wrong coordination root.

Install workspace skills with the default nested layout:

/path/to/ar-coordination/scripts/install-skills.sh \
  --install-root /path/to/openclaw-workspace/skills

For shared global skills:

/path/to/ar-coordination/scripts/install-skills.sh \
  --install-root ~/.openclaw/skills

OpenClaw resolves symlink and junction targets before loading skills. Without an explicit trust entry, it may report a symlink-escape or skip the linked runtime because the real path points outside the configured skill root. Keep the links and add a narrow allowSymlinkTargets entry to the OpenClaw config:

{
  skills: {
    load: {
      extraDirs: ["/path/to/openclaw-workspace/skills"],
      allowSymlinkTargets: ["/path/to/ar-coordination/skills"],
    },
  },
}

For a Windows workspace using .agents/skills, that looks like:

{
  skills: {
    load: {
      extraDirs: ["C:/ew/.agents/skills"],
      allowSymlinkTargets: ["C:/ew/ar-coordination/skills"],
    },
  },
}

Keep allowSymlinkTargets scoped to the real ar-coordination/skills directory. Do not allow a broad parent such as the whole workspace or home directory.

Long-running Turns

Deep reasoning and large onboarding waves can exceed OpenClaw's default patience windows. If the TUI reports that a response is taking longer than expected, or if model requests abort while DeepSeek is still thinking, increase both the agent run timeout and the provider idle timeout:

{
  agents: {
    defaults: {
      timeoutSeconds: 1800,
    },
  },
  models: {
    providers: {
      deepseek: {
        timeoutSeconds: 900,
      },
    },
  },
}

Restart the gateway after changing timeout settings:

openclaw gateway restart

You can inspect OpenClaw skill visibility with:

openclaw skills list
openclaw skills check