An agent orchestration platform with tool registry, capability management, and streaming tool execution for multi-connector workflows (HubSpot, Google Calendar, Instagram, QuickBooks, Stripe, Slack, Gmail, Google Drive, Twilio)
AgentOS exhibits severe definition quality gaps across most tools. Only 2 of 15 tools (drive.read, drive.write) have complete input schemas with property descriptions and type definitions. The remaining 13 tools (gmail.read, gmail.send, hubspot.read, hubspot.write, web.search, llm, stripe.chargeCustomer, stripe.refundPayment, shell.execute, zapier.webhook.trigger, calendar.events.list, calendar.events.get, calendar.availability.get) have empty or missing input schemas, just `{"type":"object","properties":{}}`. Descriptions exist but are minimal (10-50 chars for most), lacking WHEN to use, prerequisites, or consequences. No output schemas documented for any tool. No error handling guidance. High-risk tools (stripe.chargeCustomer, shell.execute) lack permission gates, dry-run patterns, or confirmation steps. Tool naming is inconsistent: some use verb_noun (drive.read, drive.write), others use noun.verb (calendar.events.list, gmail.read). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) present.
Check free/busy status for a set of email addresses within a date range — used to find meeting slots.
Get a single calendar event by ID from Google Calendar.
List calendar events from Google Calendar — returns events within a date range (default: next 7 days).
Read files from Google Drive — lists files and reads file content (CSV parsing for Monday loop)
Write files to Google Drive — uploads CSV or text files
Read messages from Gmail
13 of 15 tools have empty input schemas (only `{"type":"object","properties":{}}`). No parameter types, descriptions, or constraints visible.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. Violates pattern:tool and pattern:response-shaper.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 31 | 2025-06-18+ | v2 |
Send a message with Gmail
Read records from HubSpot
Write records to HubSpot
Use LLM to reason and analyze
Execute shell commands
Charge a customer via Stripe
Refund a payment via Stripe
Search the web for information
Trigger a Zapier webhook
High-risk tools (stripe.chargeCustomer, stripe.refundPayment, shell.execute) lack permission gates, dry-run patterns, confirmation steps, or error recovery guidance. No audit trail or scope declarations visible.
Descriptions are minimal (10-50 chars) and lack WHEN to use, prerequisites, or state-change consequences. E.g., 'Read messages from Gmail' does not explain filtering, pagination, or what fields are returned.
No error handling guidance. Tools do not indicate retryability, user-fixable errors, or recovery steps. LLMs cannot determine next action on failure.