Works with the AI tools you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+22 moreClaude CodeCursorCodex CLIMuseDots
OpenClaw
Manus
Grok
Grok Bot+18 agentsWebhook Failure Triage & Safe Retry
Classify failed or uncertain webhooks and return an evidence-grounded RETRY, DO_NOT_RETRY, or RECONCILE_FIRST verdict.
$7
Webhook Failure Triage & Safe Retry
Example session with this skill installed
Stripe shows that the synthetic invoice.paid event evt_demo_123 was delivered at 10:00:05 UTC. Our edge log records a 504 at 10:00:35 UTC, and the subscription dashboard still shows past_due. We do not yet have the receiver application log or an authoritative ledger lookup. The handler is intended to deduplicate on the Stripe event ID, but we have not verified the deduplication store. Is it safe to replay this event?
- Read your context and instructions
- Compiled the webhook failure triage
- Generated the document
Webhook Triage Report
VERDICT: RECONCILE_FIRST
PRIMARY_STATE: UNKNOWN
Confidence: High that the evidence is insufficient; low on the actual commit state.
The 504 shows that a gateway did not receive a timely upstream response. It does not prove whether the application rejected the event, failed before the side effect, committed the side effect and lost the acknowledgement, or never received it.
Canonical timeline
| Time (UTC) | Layer | Supplied evidence | What it proves |
|---|---|---|---|
| 10:00:05 | Provider | Stripe marks evt_demo_123 delivered | A provider-recorded delivery outcome, not the internal commit state |
| 10:00:35 | Edge | 504 recorded | A timely upstream response was not received |
| Current snapshot | Subscription view | Status remains past_due | The visible projection is not updated; an internal ledger commit is still unproven |
Evidence assessment
| Claim | Label | Assessment |
|---|---|---|
| A 504 occurred | OBSERVED | Supplied edge-log fact |
| No business side effect occurred | UNKNOWN | No authoritative ledger or receiver evidence was supplied |
| Event-ID deduplication will prevent duplication | UNKNOWN | Intended behavior is stated, but the store and atomic enforcement are unverified |
| Immediate replay is safe | UNKNOWN | Commit state and idempotency must be reconciled first |
Duplicate-impact radius
A duplicate could affect the subscription period, billing ledger, entitlement state, and customer notification. Do not use the dashboard alone as proof that none of these changed.
Minimum checks before any replay
- Query the event-processing or deduplication record for
evt_demo_123. - Check the authoritative billing ledger and subscription record for the expected side effect.
- Correlate receiver logs using the event ID and the 10:00:05–10:00:35 UTC window.
- Confirm that a replay preserves the original event identity and that deduplication is enforced atomically with the side effect.
Decision path
- If the authoritative side effect is absent and atomic deduplication is verified, the verdict may change to RETRY.
- If the side effect or processed-event record exists, use
DO_NOT_RETRY and reconcile the acknowledgement or projection instead.
- If either fact remains unavailable, keep RECONCILE_FIRST.
No replay has been executed. A human must authorize any payment-affecting action after the checks above.
webhook-failure-triage-safe-retry.pdf
PDF · document
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
What you get
About this skill
The problem
A webhook can time out or return an error after part of the business operation has already committed. Replaying too early can duplicate a payment, order, entitlement, message, or ledger entry. A 2xx response also does not prove that every downstream side effect completed.
What it does
- Classifies the supplied evidence as NOT_DELIVERED, DELIVERED_REJECTED, DELIVERED_PROCESSING_FAILED, PROCESSED_ACK_LOST, DUPLICATE, or UNKNOWN.
- Builds a canonical timeline that separates provider attempts, receiver acknowledgements, processing steps, and business side effects.
- Returns one bounded verdict: RETRY, DO_NOT_RETRY, or RECONCILE_FIRST.
- Identifies duplicate-impact radius, minimum discriminating checks, unresolved facts, and a safe retry plan when retry is justified.
What you provide
Redacted provider status, event or delivery IDs, timestamps and timezone, HTTP response or timeout evidence, receiver logs, expected side effect, current authoritative state, and verified idempotency or deduplication behavior. Never provide signing secrets, tokens, cookies, private keys, or full customer records.
What you receive
- A clear verdict and primary state with confidence.
- An evidence table labeled OBSERVED, DERIVED, or UNKNOWN.
- A canonical timeline and duplicate-impact analysis.
- The minimum checks required before any action.
- A bounded retry plan with prerequisites, verification, and stop conditions when RETRY is justified.
Why this is different
The skill does not equate a 2xx response with business completion, or a timeout or 5xx response with a safe retry. It prefers RECONCILE_FIRST when commit state or idempotency is unproven and never performs the replay itself.
Known limitations
It uses only the redacted evidence the user supplies. It does not access production systems, provider dashboards, APIs, queues, databases, files, or secrets, and it cannot prove root cause when logs are incomplete. High-impact retries require human authorization after reconciliation.
How to install
Works the same in every agent - Claude, Cursor, Codex, Copilot and 20+ more.
- 1
Download the ZIP
Free skills download straight away. Paid skills unlock right after purchase.
- 2
Unzip into your skills folder
Every agent reads skills from one folder on your machine. Drop the unzipped folder in there.
- 3
Ask your agent to use it
Restart the agent if it was already running. It picks the skill up automatically - no config needed.
Skills folder by agent
Click the path to copy it. Create the folder if it does not exist yet.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
30-day refund
Not a fit? Get your money back
Trust & safety
Security scanned
Verified clean 17 days ago
- Passed all security checks, Safe to install