Votre agent a signalé un succès.
Il n’a rien produit.

# Schedule Trigger -> HTTP Request -> AI Agent -> HTTP Request
#
# On the AI Agent node, set "On Error" to "Continue (using error output)"
# and wire that second branch to the /fail node. Otherwise a thrown tool
# error just ends the execution quietly.

HTTP Request  (before the agent — required, not optional)
    GET  https://ping.steadycron.com/abc123/start

HTTP Request  (after the agent)
    POST https://ping.steadycron.com/abc123/success
    Body (JSON):
      { "itemsProduced": {{ $json.rows.length }} }

# The /start node is required: a /success with no run open is rejected,
# because without a start time we cannot tell a slow run from a hung one.
# No IF node needed though — report the count and let SteadyCron judge it.
# An agent node returning an empty string is a perfectly successful n8n
# execution, which is exactly why that judgement belongs on our side.

# GitHub disables scheduled workflows after 60 days of repo
# inactivity, and drops schedule events under load. Neither one
# notifies you — the missing run is what tells you.

on:
  schedule:
    - cron: "0 2 * * *"
  workflow_dispatch:

jobs:
  nightly-digest:
    runs-on: ubuntu-latest
    env:
      PING: https://ping.steadycron.com/${{ secrets.SC_PING_TOKEN }}
    steps:
      - uses: actions/checkout@v4

      # Required: /success is rejected without a run open.
      - run: curl -fsS "$PING/start"

      - run: python run_agent.py --task nightly-digest

      # Report the count. A zero is a failure, and SteadyCron decides
      # that — so there is no branch here for you to get wrong.
      - run: |
          rows=$(wc -l < out/digest.md)
          body=$(printf '{"itemsProduced": %d}' "$rows")
          curl -fsS -H 'Content-Type: application/json' -d "$body" "$PING/success"

      - if: failure()
        run: |
          curl -fsS -H 'Content-Type: application/json' -d '{"status":"failure","summary":"workflow failed"}' "$PING/fail"

import requests
from my_agent import graph          # your compiled LangGraph

PING = "https://ping.steadycron.com/abc123"

def report(path="", **fields):
    # Monitoring must never be able to break the run.
    try:
        requests.post(PING + path, json=fields, timeout=5)
    except requests.RequestException:
        pass

report("/start")          # required — /success is rejected without it
try:
    state = graph.invoke({"task": "nightly-digest"})

    # Report what happened. Zero rows is a legitimate thing to report —
    # SteadyCron is what turns it into a failure and an alert.
    report(
        "/success",
        itemsProduced=len(state.get("rows", [])),
        steps=state.get("steps"),
        model="claude-opus-5",
        tokensIn=state.get("tokens_in"),
        tokensOut=state.get("tokens_out"),
        traceUrl=state.get("trace_url"),
    )
except Exception as exc:
    report("/fail", status="failure", summary=str(exc)[:500])
    raise

# One lock so runs cannot overlap, one hard deadline so a wedged
# model call cannot run until morning, one report so you find out.
0 2 * * *  flock -n /tmp/agent.lock timeout 30m /opt/agent/run.sh

# /opt/agent/run.sh
#!/usr/bin/env bash
set -uo pipefail
PING="https://ping.steadycron.com/abc123"

# Required: a /success with no run open is rejected.
curl -fsS "$PING/start"

if python run_agent.py --task nightly-digest; then
  # Report the count and let SteadyCron judge it. A zero trips the
  # empty-result rule; you do not write that branch yourself.
  rows=$(wc -l < out/digest.md 2>/dev/null || echo 0)
  body=$(printf '{"itemsProduced": %d}' "$rows")
  curl -fsS -H 'Content-Type: application/json' -d "$body" "$PING/success"
else
  curl -fsS -H 'Content-Type: application/json' -d '{"status":"failure"}' "$PING/fail"
fi