- Home
- Skills
- Agents & Orchestration
- support ticket investigation
Works with the AI Agents you already use
support ticket investigation
Cross-source support ticket investigation via Postgres, CloudWatch, and New Relic using the Investigation MCP server.
Free
support ticket investigation
Example session with this skill installed
Investigate why order 99283 failed for the customer yesterday afternoon. I have the Investigation MCP server running.
- Read your context and instructions
- Compiled the support ticket investigation
I have investigated the failure for order 99283. I queried the PostgreSQL database, correlated the request_id with CloudWatch logs, and checked for related errors in New Relic.
Investigation Report: Order 99283
| Source | Detail | Evidence |
|---|---|---|
| Postgres | Order Status | Found row in orders table. Status is FAILED. Created at 2023-10-24 14:02:10 UTC. |
| Postgres | Correlation | request_id identified as req_a1b2c3d4. user_id is 8821. |
| CloudWatch | Application Log | Found ERROR at 14:02:11 UTC in log group /aws/lambda/checkout-service. Message: ConnectionTimeoutException while calling payment gateway. |
| New Relic | System Health | TransactionError rate for checkout-service spiked to 12% between 14:00 and 14:15 UTC. |
| New Relic | Infrastructure | No significant CPU or Memory spikes found on the database or application nodes during this window. |
Analysis: The order failed due to a timeout connecting to the payment provider. This was part of a wider 15-minute spike in connection errors, suggesting a transient networking issue or downstream provider outage rather than a specific data issue with this order.
Next steps
- Check the payment provider's status dashboard for the window of 14:00-14:15 UTC yesterday.
- Run
db_run_queryto see if other orders from the same window have since succeeded or need manual reconciliation. - Use
cw_get_metric_statsto verify if the connection timeouts were limited to a specific availability zone.
Connects securely to your tools. The creator never sees your data.
About this skill
Manual troubleshooting across siloed data sources leads to slow resolution times and missed correlations. This skill provides a systematic workflow for investigating support tickets by integrating with a read-only Investigation MCP server. It connects your agent to Postgres, AWS CloudWatch, and New Relic to find root causes without manual context switching.
What it does
- Multi-source correlation helps you link database records, CloudWatch logs, and New Relic events using shared identifiers like request IDs or order IDs.
- Identifier resolution handles ambiguous store names, domains, or user IDs, ensuring you query the correct entities.
- Systematic discovery enforces a list-describe-query pattern to prevent failed queries caused by guessed schema or attribute names.
- Health monitoring retrieves error rates and infrastructure metrics to distinguish between isolated failures and systemic spikes.
- Automated guardrail handling interprets structured error messages from the MCP server to refine queries when they hit scan limits or syntax blocks.
How it works
- Fact pinning extracts identifiers and time windows from the user's ticket to set the investigation scope.
- Schema discovery runs inventory and description tools (
db_list_tables,cw_describe_log_fields) to ground the agent in your specific environment. - Evidence gathering executes targeted queries across DB, logs, and APM tools to build a chronological timeline of the failure.
- Analysis and reporting merges findings by timestamp and ID to provide a summarized evidence report.
Frameworks & tools
This skill requires a locally running Investigation MCP server. It interacts with PostgreSQL, AWS CloudWatch Logs, and New Relic NRQL using the db_*, cw_*, and nr_* tool sets.
Why this beats prompting it yourself
Writing prompts that correctly handle three different query languages while enforcing discovery-first logic is complex and prone to hallucination. This skill ensures the agent follows a rigid verification workflow and properly handles structured error feedback from the server's security guardrails.
Use cases
- Debugging failed checkout transactions using an order ID and timestamp.
- Investigating intermittent API errors by correlating Postgres statuses with CloudWatch traces.
- Checking for infrastructure resource spikes during reported periods of high latency.
- Identifying top error messages across logs to determine the scope of a customer-reported bug.
Known limitations
Requires a pre-configured Investigation MCP server with valid credentials. Does not support data modification; all queries are read-only SELECT or NRQL statements. Masked PII cannot be recovered.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Requires a self-hosted Investigation MCP server. Only supports read-only operations. Cross-source joins must be performed by the agent manually. PII is automatically redacted.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
This skill has been used 1 time.
Trust & safety
Security scan
Security scan passed 20 days ago
- Free with an Agensi account
- Software
- Opsinsight MCP
uv (Python tool runner)PostgreSQLAWS CloudWatch LogsNew RelicInvestigation MCPLog Db Investigator MCP
Python
Permissions
Allowed Hosts
All access is read-only: single SELECT queries and NRQL only, enforced by the server and a read-only DB role. Credentials stay in your own MCP client config and are never requested in chat. PII (emails, cards, phone, Aadhaar, PAN) is masked in results. CloudWatch queries have a time-window cap and a bytes-scanned cost ceiling. AWS and New Relic are optional; configure only the sources you use.