- Home
- Skills
- Testing & Debugging
- Agent Failure Detective — Trace AI Workflow Breaks
Works with the AI Agents you already use
Agent Failure Detective — Trace AI Workflow Breaks
Diagnose AI workflow failures from evidence, isolate root causes, and propose minimal fixes with regression checks.
$9.99
Agent Failure Detective — Trace AI Workflow Breaks
Example session with this skill installed
Diagnose this illustrative AI workflow incident using only the supplied evidence: a support automation times out after calling search twice, then retries the whole workflow and creates duplicate tickets. Logs show first search call succeeded in 180 ms, second returned 429 with Retry-After 2 seconds, and the ticket-create request timed out with no response. The provider and ticket API docs are not supplied. Identify facts, competing hypotheses, likely failure boundaries, safe next tests, containment, unknown outcome handling, and a minimal remediation plan. Do not assume whether the ticket was created.
- Read your context and instructions
- Compiled the agent failure detective
Incident triage: duplicate support tickets after timeout
Confirmed facts from the supplied example
- The first search call returned success in 180 ms.
- A second search call returned 429 and specified Retry-After: 2 seconds.
- The ticket-create request timed out without a response.
- The overall workflow retried, and duplicate tickets were observed.
- The provider documentation, idempotency support, and ticket-create outcome are not known.
Failure boundary and hypotheses
Rate-limit handling: The workflow may have retried search too early or restarted the whole flow after the 429. Verify the actual retry timing and whether the workflow honors the supplied delay.
2.
Write timeout ambiguity: The ticket API may have committed the first create despite the client timeout. The timeout cannot establish success or failure.
3.
Duplicate prevention gap: A workflow-level retry may issue a second create without an idempotency key or reconciliation step. Confirm using sanitized request IDs and ticket records.
These are hypotheses, not verified root causes. No provider behavior or response is assumed beyond the supplied trace.
Immediate containment
Pause automatic retries for this workflow or route uncertain creates to human review. Search for tickets matching the affected request using authorized, documented lookup methods before another create. Do not resend the timed-out write until its outcome is reconciled.
Discriminating tests
- Reproduce the 429 in a sandbox and verify the workflow waits at least the documented Retry-After duration without restarting successful steps.
- Simulate a timeout after server-side commit and determine whether an idempotency key or status/lookup endpoint can identify the original ticket.
- Trace the workflow’s retry boundary, request IDs, and ticket-create invocations to distinguish duplicate orchestration from duplicate provider behavior.
- Test a definitive non-committing failure and a malformed response separately.
Minimal remediation plan
- Retry only the failed read operation after bounded backoff; preserve successful prior search results when safe.
- Give each intended ticket creation a stable idempotency key if the API documents support. Otherwise, reconcile through an authorized lookup or stop for human review after an ambiguous timeout.
- Represent write outcomes as confirmed, rejected, pending, or unknown; never treat a timeout as proof of failure.
- Add a regression test for 429 handling, timeout-after-commit, and repeated workflow execution. Measure duplicate-create rate and alert on ambiguous writes.
Required evidence
Obtain provider documentation, sanitized traces with timestamps and request IDs, workflow retry configuration, ticket audit history, idempotency behavior, and a sandbox reproduction before naming a confirmed root cause.
Connects securely to your tools. The creator never sees your data.
About this skill
Find the cause behind an AI workflow failure
Agent Failure Detective investigates agents, automations, APIs, MCP tools, RAG, memory, queues, webhooks, databases, browser workflows, schedulers, model providers, and multi-step systems. It builds a timeline from user-supplied logs, traces, code, screenshots, schemas, prompts, test results, and incident notes; identifies the failure boundary; compares competing hypotheses; proposes discriminating tests; traces propagation; and recommends the smallest corrective change with verification and regression guards.
The skill separates observed facts, inferred explanations, and unknowns. It preserves working behavior, avoids symptom-only patches, and uses live tools only when actually available and authorized. It never claims to have inspected, reproduced, executed, queried, or verified evidence it has not accessed. It never fabricates logs, request IDs, responses, provider behavior, timestamps, test results, root causes, or successful fixes.
Best results
Share the exact symptom, expected behavior, recent changes, timestamped sanitized logs, request/trace IDs, relevant code/config/schema, environment, reproduction steps, and business impact. Remove credentials, tokens, customer data, and unrelated sensitive details.
A root cause cannot be confirmed when the evidence is incomplete, ambiguous, or non-reproducible. Logs may omit events or use unsynchronized clocks. Provider and connector behavior requires current documentation or verified tests. Suggested commands and changes must be reviewed for environment and security impact before use.
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 any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 month ago
Permissions
Read only user-authorized code, logs, traces, screenshots, schemas, and incident evidence. Terminal or shell, browser, and network use only when available, necessary, and authorized. Do not query private systems or execute remediation without explicit permission. Never request or expose secrets; redact sensitive data.