OpenClaw Agent Feishu Onboarding

Standardize creation of new OpenClaw agents and Feishu route binding.

Use this skill for operational execution with strict confirmations.

Runtime Prerequisites

• Required binaries:
• openclaw (CLI available in PATH)
• python (3.x, used by scripts/validate_feishu_bindings.py)
• Preflight checks:
• openclaw --version
• python --version
• Data access behavior:
• Reads local OpenClaw config (openclaw.json / OPENCLAW_CONFIG_PATH)
• May write local agent and routing state (agents.list[], bindings[])
• Do not run on machines where local config access is not permitted

Core Scope

• Create a new agent (agents.list + workspace + identity).
• Confirm Feishu route target fields before any write.
• Bind routing using bindings.match with channel/accountId/peer.
• Verify final route and provide rollback steps.

Required Inputs

• Agent intent:
• Agent purpose and responsibility boundary.
• Explicit out-of-scope items.
• Agent spec:
• agentId (lowercase, digits, hyphen).
• workspace path.
• model id.
• Identity object in agents.list[]:
• use identity object (for example identity.name, identity.emoji, identity.theme)
• do not rely on top-level name only
• Feishu route spec:
• match.channel: feishu
• match.accountId: for example main
• match.peer.kind: group (or dm when needed)
• match.peer.id: Feishu session id (group id like oc_xxx or dm id)

Hard Validation Gates (Must Pass)

• agents.list[] entry for new agent must include identity object.
• identity.name must be set for human-readable routing/debug checks.
• match.accountId must be one of channels.feishu.accounts keys (for example main).
• Never put Feishu group/session id (oc_xxx) into match.accountId.
• Group routing must include match.peer = { kind: "group", id: "oc_xxx" }.
• If routing is for a specific group but match.peer is missing, abort and correct config before continuing.
• Before write, print the final binding object and require explicit confirmation.

Multi-Step Confirmation Protocol

Do not skip confirmations. Ask and confirm in this exact order.

1. Confirmation A: Agent Goal
• Confirm what this agent should do.
• Confirm what this agent must not do.
2. Confirmation B: Agent Configuration
• Confirm agentId, workspace, model, identity fields.
• Confirm naming conventions and collision check (agentId uniqueness).
3. Confirmation C: Feishu Routing Target
• Confirm accountId.
• Confirm peer.kind.
• Confirm peer.id (explicitly state this is the Feishu session id).
• Confirm whether this is a precise peer binding or account-level fallback.
4. Confirmation D: Execution Approval
• Summarize all fields in one compact block.
• Ask for final go/no-go before writing config.

Execution Workflow

1. Discover current state:
• openclaw agents list
• openclaw config get agents.list --json
• openclaw config get channels.feishu.accounts --json
• openclaw directory groups list --channel feishu --account --query "" --json
• openclaw config get bindings --json
2. Create agent:
• openclaw agents add ...
• Optional: openclaw agents set-identity ...
3. Apply routing:
• Use openclaw agents bind for account-scoped binding when needed.
• Ensure peer-precise rule exists in top-level bindings[] with match.peer.
4. Validate:
• python -X utf8 ./scripts/validate_feishu_bindings.py --config
• openclaw config validate --json
• ensure target agents.list[] entry has identity.name
• Check bindings[] entry content and ordering.
• Ensure every group-targeted rule has match.peer.kind = "group" and match.peer.id = "oc_xxx".
5. Reload/restart gateway if required by deployment policy.

Routing Rules

• peer rule is the precise route key for a specific Feishu conversation.
• Use match.peer.id to bind a specific session (group oc_xxx).
• If both peer-level and account-level bindings exist, keep peer rule first.
• Avoid broad fallback rules until peer-specific routes are confirmed.
• Wrong example (forbidden): match.accountId: "oc_xxx" without match.peer.

Agent list entry should follow this shape:

{
  "id": "data-analyst",
  "workspace": "C:\\Users\\Administrator\\.openclaw\\workspace-dataAnalysis",
  "agentDir": "C:\\Users\\Administrator\\.openclaw\\agents\\data-analyst\\agent",
  "identity": {
    "name": "data-analyst"
  }
}

Canonical peer binding object:

{
  "agentId": "<agent-id>",
  "match": {
    "channel": "feishu",
    "accountId": "main",
    "peer": {
      "kind": "group",
      "id": "oc_xxx"
    }
  }
}

Safety and Rollback

• Never delete agents unless explicitly requested.
• Before changing bindings, capture current bindings snapshot.
• If wrong route is applied, revert to previous binding set and re-validate.
• Prefer reversible, explicit changes and post-change verification.

Why It Still Routes to main

• Binding did not match the real incoming peer.id (most common).
• bindings[] was written but gateway runtime did not reload the latest config.
• match.accountId mismatched current Feishu account key.
• Target agent exists but agents.list[] entry is malformed (missing required fields such as identity object expected by your convention).
• Message arrived in a different conversation than the configured group.

References

• Command examples and rollback snippets: commands.md
• Chinese runbook and confirmation checklist: usage-zh.md
• Auto-check script: validate_feishu_bindings.py