supplygraph-ai

Corporate Exception Report Agent

Overview

The Corporate Exception Report Agent continuously monitors and analyzes companies to detect critical exceptions, risk signals, regulatory events, and abnormal changes that can impact operational stability, compliance posture, or reputation.

It transforms fragmented, unstructured external information into a structured, verified, and continuously updated intelligence report, enabling organizations to respond to corporate risks in near real-time.

Pain

Traditional corporate monitoring is slow, fragmented, and resource-intensive.

Analysts frequently spend 10+ hours per company gathering information from scattered public sources — and by the time a report is completed, much of the data is already outdated.

Most organizations also lack:

This leads to blind spots in:

Breakthrough

The Corporate Exception Report Agent automates what previously required multiple teams and systems.

Behind the scenes, it:

For the user, the workflow remains simple:

  1. Provide a company name
  2. Confirm the matched entity
  3. Receive a detailed, structured exception report

Typical insights include:

Why SupplyGraph AI

Powered by:

SupplyGraph AI provides a continuously updated, audit-ready corporate exception intelligence layer without requiring any proprietary data.

This makes the agent suitable for:

Agent Behavior Model

Sandbox Key Support (for Development)

This agent fully supports Sandbox API Keys, enabling developers to test integrations without consuming credits.

When using a Sandbox Key:

Sandbox Keys are ideal for:

⚠️ Sandbox data must not be used for production analytics.

The full A2A workflow (run → status → results) behaves the same as production, enabling accurate integration testing.

Initial Data Preparation (First-Time Subscription)

When a user initiates a new monitoring request (via /run without a task_id), the agent begins constructing a full baseline dataset for the specified company.

This includes:

This initial build requires approximately 2–3 hours.

First-time workflow:

  1. User calls /run without task_id → system creates task_id
  2. User confirms the matched company
  3. Agent begins data preparation
  4. User polls the Status Endpoint
  5. Once TASK_COMPLETED is returned → first dataset is ready
  6. User retrieves the initial report via mode=results

After this initial build:

Subscription & Billing Model

When a user confirms monitoring for a specific company:

A subscription remains active as long as:

Understanding task_id and Subscription Lifecycle

A task_id represents a subscription instance, not a company identity.

How task_id is created and maintained

Implications:

Results Endpoint Behavior During Subscription

The results endpoint always returns the current latest snapshot for the monitored company, as long as the task_id is active.

Subscription Cancellation

A subscription may end through:

1. User-Initiated Cancellation

Users may cancel by calling /run with mode=cancel + task_id.

Example: Cancel Subscription

curl -X POST https://agent.supplygraph.ai/api/v1/agents/corporate_exception_report/run \
  -H "Authorization: Bearer <YOUR_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
        "mode": "cancel",
        "task_id": "<task-id>"
      }'
Example Response
{
  "success": true,
  "code": "TASK_CANCELLED",
  "data": {
    "task_id": "<task-id>",
    "agent": "corporate_exception_report",
    "stage": "cancelled",
    "content": "Your subscription has been successfully cancelled. Monitoring will continue until the current billing period ends."
  }
}

2. Credit Deduction Failure

If credits cannot be deducted:

Any restart requires a new subscription, resulting in a new task_id.

Input Requirements

agent_id: corporate_exception_report · MCP tool: corporate_exception_report

Field Required Description
text Yes Company name and country
chapter_name No Exception domain; default ALL

chapter_name values: ALL, Profile, License, GovRel, R&D, Reputation, Ops, GeneralRisks, Cost, Compliance, Competition, Brand, HR

Supported modes: run, status (initial build only), results, cancel. See Agent Behavior Model above for subscription lifecycle.

Output Format

Primary output lives in data.content (Agent API) or task artifacts (A2A / MCP). The payload is a oneOf:

State content type Description
In progress, failed, cancelled, subscription inactive, or waiting for user input string Text or Markdown — validation prompts, confirmation requests, or error messages
Task completed successfully object Structured corporate exception report (see below)

Structured success object:

{
  "type": "results",
  "data": {
    "languages": ["en", "zk"],
    "report_infos": {
      "en": {
        "report_date": "2025-11-30",
        "report_name": "Corporate Exception Report — Tesla, Inc.",
        "report_type": "enterprise changes report"
      },
      "zh": {
        "report_date": "2025-11-30",
        "report_name": "企业异常报告 — Tesla, Inc.",
        "report_type": "enterprise changes report"
      }
    },
    "chapter_infos": [
      {
        "en": {
          "chapter_name": "Compliance",
          "sub_sections": [],
          "chapter_texts": ["Markdown paragraph…"]
        },
        "zh": {
          "chapter_name": "合规风险",
          "sub_sections": [],
          "chapter_texts": ["Markdown 段落…"]
        }
      }
    ]
  }
}
Field Description
type Always "results" on success
data.languages Language codes in the response — en (English), zk (Chinese)
data.report_infos Report metadata keyed by language (en required; zh when Chinese is included)
data.chapter_infos[] Array of exception-domain chapters, each with en / zh blocks
chapter_infos[].en\|zh.chapter_name Chapter title (e.g. Compliance, Reputation, Ops)
chapter_infos[].en\|zh.sub_sections[] Nested sub-sections (section_title, sub_sections, section_texts)
chapter_infos[].en\|zh.chapter_texts[] Chapter body paragraphs (Markdown)

During an active subscription, each results call returns the latest snapshot of the monitored company — not a one-time static export.

Initial baseline build: ~2–3 hours (Production). Sandbox returns instantly.

Sample Response (Sandbox)

Sandbox returns Tesla, Inc. or BYD fixture data with the same structure as Production. Content below is abbreviated.

Success (TASK_COMPLETED) — Agent API results:

{
  "success": true,
  "code": "TASK_COMPLETED",
  "message": "Task completed successfully.",
  "data": {
    "task_id": "<task-id>",
    "agent": "corporate_exception_report",
    "stage": "completed",
    "progress": 100,
    "content": {
      "type": "results",
      "data": {
        "languages": ["en", "zk"],
        "report_infos": {
          "en": {
            "report_date": "2025-11-30",
            "report_name": "Corporate Exception Report — Tesla, Inc.",
            "report_type": "enterprise changes report"
          },
          "zh": {
            "report_date": "2025-11-30",
            "report_name": "企业异常报告 — Tesla, Inc.",
            "report_type": "enterprise changes report"
          }
        },
        "chapter_infos": [
          {
            "en": {
              "chapter_name": "Compliance",
              "sub_sections": [
                {
                  "section_title": "Regulatory Actions",
                  "sub_sections": [],
                  "section_texts": ["No material regulatory enforcement actions detected in the monitoring window."]
                }
              ],
              "chapter_texts": ["### Summary\nCompliance posture remains stable with no critical exceptions flagged."]
            },
            "zh": {
              "chapter_name": "合规风险",
              "sub_sections": [],
              "chapter_texts": ["### 摘要\n监控周期内未发现重大合规异常信号。"]
            }
          },
          {
            "en": {
              "chapter_name": "Reputation",
              "sub_sections": [],
              "chapter_texts": ["### Media Exposure\nNegative media sentiment remains within normal baseline levels."]
            },
            "zh": {
              "chapter_name": "声誉风险",
              "sub_sections": [],
              "chapter_texts": ["### 媒体曝光\n负面舆情处于正常基线范围内。"]
            }
          }
        ]
      }
    }
  },
  "metadata": { "credits_used": 0 },
  "errors": null
}

Multi-turn Example (A2A)

When the agent cannot unambiguously match a company from text, it returns WAITING_USER with a Markdown prompt in content (string). Continue with the same task_id.

Turn 1 — submit company name (optional chapter_name):

curl -X POST "https://agent.supplygraph.ai/api/v1/agents/corporate_exception_report/run" \
  -H "Authorization: Bearer <YOUR_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{"text": "Tesla, Inc. United States", "chapter_name": "ALL", "stream": false}'

Poll until WAITING_USER. Example content:

{
  "success": true,
  "code": "WAITING_USER",
  "data": {
    "task_id": "<task-id>",
    "content": "Here are the companies we've identified.\nCompany Name: Tesla, Inc.\nCountry: United States\nPlease reply [Yes] or [No] to confirm."
  }
}

Turn 2 — confirm entity (same task_id):

curl -X POST "https://agent.supplygraph.ai/api/v1/agents/corporate_exception_report/run" \
  -H "Authorization: Bearer <YOUR_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{"text": "yes", "task_id": "<task-id>", "stream": false}'

The agent activates the subscription and begins baseline data preparation. Poll status every ~10 seconds until TASK_COMPLETED (initial build ~2–3 h in Production), then retrieve the report via results.

After the initial build completes, use results only to fetch the latest snapshot — no further status polling is required during the subscription.

See Agent API §8 for the general multi-turn pattern.

Integration

Method ID / Tool Documentation
A2A corporate_exception_report a2a.md
MCP corporate_exception_report mcp.md
Agent API corporate_exception_report agent-api.md

Quick examples: A2A / MCP · Agent API

Errors

Common codes → Agent API §10. Supports mode=cancel for subscription cancellation. Returns WAITING_USER for company confirmation.

Maintainer: info@supplygraph.ai
License: Proprietary / Internal
© 2025–2026 SupplyGraph AI. All rights reserved.