OpenClaw AICFO Agent

Overview

Use this skill to connect OpenClaw to the AICFO platform without touching browser-only routes or tenant-local daemons.

Primary surfaces:

• GET /api/agent/session
• GET /api/agent/companies
• GET /api/agent/dashboard
• GET|POST /api/agent/connectors
• GET /api/company/query
• GET /api/company/search
• GET /api/company/entity
• GET /api/company/file
• GET /api/documents
• GET /api/documents/questions
• POST /api/documents/:id/clarifications

Canonical references:

• docs/contracts/openclaw-quickstart.md
• docs/contracts/external-agent-access.md

Quick Start

1. Introspect the key:
• node bin/openclaw-aicfo-adapter.mjs session
2. List available tools:
• node bin/openclaw-aicfo-adapter.mjs tools
3. Resolve company scope:
• use implicitCompanyId if present
• otherwise choose a company from companyAccess.companies
• pass the company UUID when available; the API also accepts slug or exact company name as company_id
4. Run business operations through the adapter:
• dashboard
• query/search
• get-entity
• get-file
• connector
• documents
• document-questions
• answer-document-questions

Workflow

1. Bootstrap

Always start with session.

Read:

• granted scopes
• implicitCompanyId
• requiresCompanySelection
• accessible companies

If there is no implicit company, require OpenClaw to pass company_id on every operation. Prefer the UUID from companyAccess.companies[].id; slug or exact company name are accepted as fallbacks.

2. Prefer the published adapter

Use bin/openclaw-aicfo-adapter.mjs as the primary OpenClaw bridge.

It is intentionally REST-first so the published skill does not depend on local npm packages or preinstalled MCP SDK modules.

Do not use /api/company/mcp as the primary integration path for OpenClaw.

3. Use REST only when it is simpler

REST is acceptable for:

• /api/agent/session
• /api/agent/connectors if a client wants plain JSON instead of MCP

Do not use browser onboarding routes under /api/connections/*.

Common Operations

Load references/operations.md when you need concrete command envelopes.

Typical adapter calls:

• session
• tools
• companies
• dashboard
• query
• search
• get-entity
• get-file
• connectors
• connector
• documents
• document-questions
• answer-document-questions

The published skill package is self-contained and should run on a machine that only has Node and the installed skill files.

Guardrails

• Treat /api/agent/connectors as the canonical machine connector surface.
• Treat /api/connectors/hub as legacy Slack/Jira compatibility only.
• Prefer company_id from session introspection instead of guessing. Use slug or exact company name only when the UUID is not available.
• If a key lacks documents.write, do not trigger Google Drive imports or document deletes.
• If a key uploads or imports documents, check document-questions before assuming processing is complete.
• Answer clarification requests with answer-document-questions and include provenance/source notes when available.
• If a key lacks a connector-use scope, do not attempt that provider action.
• Do not assume browser-managed integrations can be onboarded by OpenClaw.

Validation

Before relying on the integration:

1. run node bin/openclaw-aicfo-adapter.mjs session
2. run node bin/openclaw-aicfo-adapter.mjs tools
3. run one read operation against a real company:
• dashboard
• or query
4. if connector access is needed, run one bounded action such as:
• telegram list_chats
• google_drive list_files