How do I install the Silent Failure Detector skill?

Run npx skillmds@latest add alidoesai/silent-failure-detector in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Silent Failure Detector safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Silent Failure Detector?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Silent Failure Detector free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under Apache-2.

Who published Silent Failure Detector?

alidoesAi (@alidoesai) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Silent Failure Detector by alidoesai

Name: Silent Failure Detector
Author: alidoesAi

Silent Failure Detector

Detects when a heartbeat or log stream has gone SILENT unexpectedly — the failure mode where a process stops emitting without raising an error, so error-based monitoring misses it entirely. Use to verify a scheduled job, agent, logger, cron task, or data feed is still actually running, or to audit a list of timestamps for an unexpected stop. Triggers: "did this job stop", "silent failure", "stale heartbeat", "is my logger still running", "no errors but no output", "detect missing data". Deterministic, read-only, stdlib-only; ships with a reproducible eval (BENCHMARK.md) that proves its accuracy AND its limits.

alidoesAi ab67d16 11 files · 24.9 KB Updated Sep 17, 2026

File contents

silent-failure-detector

Catches the failure that monitoring-on-errors can't see: a process that just goes quiet. No exception, no crash — it simply stops emitting, and you find out days later.

When to use

You have a stream of emission timestamps (a logger, heartbeat, cron job, agent loop, data feed) and an expected cadence, and you want to know whether it has stopped.
You want a verifiable, deterministic check you can wire into a daily integrity sweep.

How it works (one knob)

SILENT when the most-recent signal is older than expected_interval_sec * multiplier (default multiplier = 3.0 → roughly three missed beats). An empty stream is SILENT. The multiplier trades detection latency (lower = catch a real death faster) against false alarms on bursty cadences (higher = more tolerant). Set expected_interval_sec to the stream's typical gap; if the cadence is bursty, set it nearer the p95 gap and say so.

Run it

echo '{"now": 1750000000, "expected_interval_sec": 60,
       "timestamps": [1749999940, 1749999990]}' > input.json
python3 detector.py input.json     # exit 0 = healthy, exit 1 = silent (alertable)

Or import it: from detector import detect; detect(now, timestamps, expected_interval_sec, multiplier=3.0).

Output

{ "status": "silent", "reason": "last signal 19880s old vs threshold 180s",
  "last_age_sec": 19880, "threshold_sec": 180.0 }

Known limits (measured, not claimed — see BENCHMARK.md)

On a 105-case labeled set it scores 83.8% accuracy (TPR 81.8% / TNR 86.0%). Every error is in the 2–4× "grey zone" near the threshold; clear cases are 100% correct. Two honest failure modes: (1) false alarms on alive-but-currently-slow streams that crossed 3× by chance; (2) a one-window detection latency on a stream that just died. Tune multiplier for your latency/false-alarm preference. The eval is reproducible — re-run it and check the numbers yourself.

---name: silent-failure-detectordescription: Detects when a heartbeat or log stream has gone SILENT unexpectedly — the failure mode where a process stops emitting without raising an error, so error-based monitoring misses it entirely. Use to verify a scheduled job, agent, logger, cron task, or data feed is still actually running, or to audit a list of timestamps for an unexpected stop. Triggers: "did this job stop", "silent failure", "stale heartbeat", "is my logger still running", "no errors but no output", "detect missing data". Deterministic, read-only, stdlib-only; ships with a reproducible eval (BENCHMARK.md) that proves its accuracy AND its limits.license: Apache-2.0---# silent-failure-detectorCatches the failure that monitoring-on-errors can't see: a process that just **goes quiet**. No exception, no crash — it simply stops emitting, and you find out days later.## When to use- You have a stream of emission timestamps (a logger, heartbeat, cron job, agent loop, data feed) and an expected cadence, and you want to know whether it has stopped.- You want a verifiable, deterministic check you can wire into a daily integrity sweep.## How it works (one knob)SILENT when the most-recent signal is older than `expected_interval_sec * multiplier` (default `multiplier = 3.0` → roughly three missed beats). An empty stream is SILENT. The multiplier trades **detection latency** (lower = catch a real death faster) against **false alarms** on bursty cadences (higher = more tolerant). Set `expected_interval_sec` to the stream's typical gap; if the cadence is bursty, set it nearer the p95 gap and say so.## Run it```bashecho '{"now": 1750000000, "expected_interval_sec": 60, "timestamps": [1749999940, 1749999990]}' > input.jsonpython3 detector.py input.json # exit 0 = healthy, exit 1 = silent (alertable)```Or import it: `from detector import detect; detect(now, timestamps, expected_interval_sec, multiplier=3.0)`.## Output```json{ "status": "silent", "reason": "last signal 19880s old vs threshold 180s", "last_age_sec": 19880, "threshold_sec": 180.0 }```## Known limits (measured, not claimed — see BENCHMARK.md)On a 105-case labeled set it scores **83.8% accuracy (TPR 81.8% / TNR 86.0%)**. **Every error is in the 2–4× "grey zone" near the threshold; clear cases are 100% correct.** Two honest failure modes: (1) false alarms on alive-but-currently-slow streams that crossed 3× by chance; (2) a one-window detection latency on a stream that *just* died. Tune `multiplier` for your latency/false-alarm preference. The eval is reproducible — re-run it and check the numbers yourself.

Frequently asked questions