Operate a Podium webhook receiver that survives the delivery-side failures — forged events without signature verification, replay attacks against a stateless handler, duplicate processing from Podium's 24h retry policy, lost events with no dead-letter queue, out-of-order batch deliveries, and timing-attack-vulnerable HMAC compares. Use when building a webhook endpoint for call transcripts, webchat events, conversation lifecycle, or review notifications; hardening an existing handler that processes events twice or drops them silently; or wiring a DLQ + replay path before the on-call rotation starts. Trigger with "podium webhook", "podium hmac", "podium signature", "podium webhook idempotency", "podium webhook replay", "podium dlq", "podium webhook retries".
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
Operate a Podium webhook receiver that survives the delivery-side failures — forged events without signature verification, replay attacks against a stateless handler, duplicate processing from Podium's 24h retry policy, lost events with no dead-letter queue, out-of-order batch deliveries, and timing-attack-vulnerable HMAC compares. Use when building a webhook endpoint for call transcripts, webchat events, conversation lifecycle, or review notifications; hardening an existing handler that processes events twice or drops them silently; or wiring a DLQ + replay path before the on-call rotation starts. Trigger with "podium webhook", "podium hmac", "podium signature", "podium webhook idempotency", "podium webhook replay", "podium dlq", "podium webhook retries".
Receive Podium webhooks in production without forged events, double-charged AI side-effects, lost notifications, or out-of-order conversation events. This is not an introductory webhook walkthrough — it is the receiver code your integration runs when Podium retries a 5xx response six times over 24 hours, when a leaked secret lets an attacker POST forged events, when a batch delivery arrives with conversation.deleted ahead of conversation.created, and when on-call needs to drain and replay 800 failed events without re-firing the ones that already succeeded.
The six production failures this skill prevents:
Missing signature verification — a webhook endpoint that accepts any POST will accept forged events. An attacker who learns the URL can create phantom contacts, fire phantom review requests, or impersonate a real customer in a webchat. HMAC-SHA256 over the raw request body is non-optional and must run before any handler logic.
Replay attacks against a stateless handler — a valid signed event POSTed twice (or 1000 times) re-runs every side effect each time. Signature validity alone is not enough — the receiver must reject events whose timestamp falls outside a 5-minute window AND whose nonce has already been seen.
Duplicate event processing from Podium retries — Podium retries webhook delivery on 5xx for up to 24 hours. Without an idempotency cache, every retry re-runs the handler (writes the contact again, fires the review request again, double-charges an AI call). SET NX EX 86400 on the event_id is the cheapest fix that exists.
Lost events without a dead-letter queue — if a handler raises and Podium retries six times and gives up, the event is gone. On-call has nothing to replay. Every handler exception must persist the raw signed payload to a DLQ before the response returns 5xx, so the event is recoverable independent of Podium's retry clock.
Batch event reordering — Podium can deliver multiple events in one POST and ordering across deliveries is not guaranteed. A naive handler processes conversation.deleted before conversation.created and the system observes a delete on a contact that does not exist. Within a batch, sort by occurred_at before dispatch; across batches, gate causally-dependent handlers on the precondition existing.
Timing-attack vulnerability on signature compare — received_sig == computed_sig with == short-circuits on the first byte mismatch. An attacker measures response latency to recover the signature byte-by-byte over a few thousand probes. Always use , which is constant-time over the longer of the two inputs.
hmac.compare_digest
Prerequisites
Python 3.10+ with fastapi, uvicorn, httpx, and redis (in-memory fallback for dev is provided)
Podium account with an OAuth app authorized for webhook delivery: Settings → Developer → Apps → Webhooks
The webhook signing secret from the app's Webhooks tab (saved to a secret store — never committed)
A receiver URL reachable from Podium (publicly resolvable HTTPS endpoint with valid cert)
Redis 6+ for production dedup + DLQ; an in-memory dict + SQLite file fallback exists for dev
A podium-auth instance if your handler needs to call back into the Podium API after processing
Instructions
Build in this order. Each section neutralizes one production failure mode.
1. HMAC-SHA256 signature verification on the raw body (neutralizes forgery)
Verify the signature against the raw, unparsed request body. Any framework middleware that JSON-decodes-and-re-encodes before signature check will fail because whitespace and key ordering change. Read the body once, verify, then parse:
import hmac, hashlib
from fastapi import FastAPI, Request, HTTPException, Header
app = FastAPI()
SIGNING_SECRET = os.environ["PODIUM_WEBHOOK_SECRET"].encode("utf-8")
@app.post("/webhooks/podium")asyncdefreceive(request: Request, x_podium_signature: str = Header(None)):
raw = await request.body() # bytes — DO NOT decode/re-encodeifnot x_podium_signature:
raise HTTPException(401, "missing X-Podium-Signature")
ifnot verify_signature(raw, x_podium_signature):
raise HTTPException(401, "signature mismatch")
# ... continue with replay/dedup/dispatch
defverify_signature(body: bytes, header_value: str) -> bool:
# Podium signature header format: "t=<unix_ts>,v1=<hex_hmac>"# Adapt to current spec — verify against the Podium developer docs at integration time.
parts = dict(p.split("=", 1) for p in header_value.split(",") if"="in p)
ts, sig = parts.get("t"), parts.get("v1")
ifnot ts ornot sig:
returnFalse
signed_payload = f"{ts}.".encode("utf-8") + body
expected = hmac.new(SIGNING_SECRET, signed_payload, hashlib.sha256).hexdigest()
return hmac.compare_digest(expected, sig) # constant-time, byte-by-byte safe
The t= timestamp is what makes the next mitigation possible. A signature alone with no timestamp is replayable forever.
Reject any event whose signed timestamp is more than 5 minutes from now (in either direction — clock skew goes both ways). This bounds the replay window an attacker has even if they capture a valid signed event off the wire:
import time
REPLAY_WINDOW_SECONDS = 300# 5 minutes; tune to your clock-skew tolerancedefwithin_replay_window(ts_str: str) -> bool:
try:
ts = int(ts_str)
except (TypeError, ValueError):
returnFalsereturnabs(time.time() - ts) <= REPLAY_WINDOW_SECONDS
Wire within_replay_window(parts["t"]) immediately after signature verification. A failed window check is a 401 — do not return 200, do not enqueue, do not log the body (the attacker is probing).
3. Idempotent dedup with SET NX EX 86400 (neutralizes duplicate processing)
Every Podium webhook carries an event_id (or equivalent unique identifier — verify against the current schema). Reject any event whose event_id is already in the dedup cache. Use Redis SET key value NX EX 86400 so the check and the claim are atomic; 86400 seconds matches Podium's 24-hour retry ceiling:
import redis.asyncio as redis
REDIS = redis.from_url(os.environ.get("REDIS_URL", "redis://localhost:6379/0"))
asyncdefclaim_event(event_id: str) -> bool:
# Returns True if this process is the first to see this event_id.# Returns False if the event_id is already in the cache (duplicate).returnawait REDIS.set(f"podium:evt:{event_id}", "1", nx=True, ex=86400)
Returning 200 on duplicate is correct — Podium has correctly delivered, the receiver has correctly identified it as already processed. The handler is idempotent by construction.
For dev / smoke environments without Redis, fall back to an in-memory set() with a periodic eviction loop. Documented in references/implementation.md.
4. Dead-letter queue before responding 5xx (neutralizes silent event loss)
Wrap every handler invocation in a try/except. On any exception, persist the raw signed payload plus the timestamp plus the signature to the DLQ before letting the exception bubble. The DLQ entry is the recovery anchor — dlq_replay.py can re-POST it to the handler later:
Redis list LPUSH podium:dlq + scheduled archiver to S3/GCS
Default for prod
SQLite file at /var/lib/podium-dlq.sqlite
Single-node deployments, dev
Append-only JSONL at /var/log/podium-dlq.jsonl
Fallback when nothing else is available — durable, parseable, ugly
The DLQ is durable independent of the Redis dedup cache. If Redis dies, dedup is degraded but events are still recoverable.
5. Batch event ordering by occurred_at (neutralizes reordering)
Podium can deliver multiple events in one POST. Within the batch, sort by occurred_at ascending before dispatch. Across batches, do not assume earlier-timestamped events arrived first — guard causally-dependent handlers with an existence check:
asyncdefdispatch_batch(events: list[dict]):
events.sort(key=lambda e: (e.get("occurred_at", 0), e.get("id", "")))
for event in events:
await safe_dispatch_one(event)
asyncdefhandle_conversation_deleted(event: dict):
convo_id = event["data"]["conversation_id"]
# Guard: if the create event hasn't been processed yet, defer this delete.ifnotawait convo_exists(convo_id):
await dlq_persist({
"reason": "out_of_order_delete_before_create",
"event_id": event["id"],
"raw_body": json.dumps(event),
"received_at": time.time(),
})
returnawait delete_conversation_locally(convo_id)
Sorting within a batch is cheap and correct. Cross-batch ordering is undecidable from the receiver side — the DLQ + replay path is the recovery mechanism when out-of-order delivery violates a precondition.
The single most common implementation bug in webhook receivers is received == expected with ==. Python string == short-circuits on the first differing byte; an attacker measures response latency over a few thousand probes and reconstructs the signature byte by byte.
# WRONG — leaks signature byte-by-byte via timingif received_sig == expected_sig:
returnTrue# CORRECT — constant-time over the longer of the two inputsif hmac.compare_digest(received_sig, expected_sig):
returnTrue
hmac.compare_digest is the only acceptable comparison. The same rule applies to Node (crypto.timingSafeEqual), Go (hmac.Equal), and Rust (subtle::ConstantTimeEq).
Podium does NOT retry — investigate Podium-side payload
200 OK (duplicate)
event_id already in dedup cache
Podium stops retrying — system is idempotent
200 OK (processed)
Handler dispatched successfully
Podium stops retrying — normal path
200 OK (deferred)
Out-of-order event written to DLQ; will resolve via replay
Podium stops retrying — recovery is internal
500 Internal Server Error
Handler raised; DLQ entry persisted
Podium retries with exponential backoff up to 24h
503 Service Unavailable
Redis dedup unreachable; handler refuses
Podium retries — fail-closed is the safe default
Examples
Verify a captured webhook payload from the command line
# Use the CLI bundled with the skill to verify a captured payload + header against the secret.
python3 scripts/signature_verify.py \
--body-file /tmp/captured_webhook_body.json \
--signature-header "t={your-timestamp},v1={your-podium-signature}" \
--secret-env PODIUM_WEBHOOK_SECRET
# exit 0 = valid; exit 1 = signature mismatch; exit 2 = replay window exceeded
Manually check if an event_id has been seen
python3 scripts/dedup_check.py --event-id evt_{your-event-identifier} --redis-url redis://localhost:6379/0
# exit 0 = first sight (would be processed); exit 1 = duplicate (would be rejected)
Drain the DLQ and replay events through the handler
# After a handler bug is fixed, replay DLQ entries through the receiver.# The replay path goes through the SAME endpoint as Podium, so signature + dedup still apply.
python3 scripts/dlq_replay.py \
--target-url https://your-receiver.example.com/webhooks/podium \
--secret-env PODIUM_WEBHOOK_SECRET \
--batch-size 25 \
--rate-per-sec 10
The replay script reuses the original signature header captured at DLQ-persist time — Podium's signing secret is the same secret your replayer uses to compute the header, so no re-signing is required for events captured within the secret's lifetime.
Boot the receiver locally for development
export PODIUM_WEBHOOK_SECRET={your-webhook-secret}
export REDIS_URL=redis://localhost:6379/0 # or unset to use in-memory fallback
uvicorn scripts.webhook_server:app --host 0.0.0.0 --port 8080 --reload
Output
FastAPI receiver with HMAC verification on the raw body, replay window, dedup, DLQ, and batch ordering
Signature verifier CLI (signature_verify.py) for incident forensics on captured payloads
Dedup-cache checker CLI (dedup_check.py) for confirming a specific event was already processed
DLQ replayer CLI (dlq_replay.py) for draining persisted failures after a handler fix
Redis-backed dedup with 24h TTL aligned to Podium's retry ceiling
DLQ persistence with multiple backend options (Redis list, SQLite, JSONL)
.gitignore rules covering the webhook secret + captured payload files
