One post tagged with "hermes"

Aturan kontes: ≤90 detik, harus memperlihatkan plugin benar-benar bekerja, tag @NousResearch di post. Judge tidak menonton 90 detik dua kali — 3 detik pertama menentukan.

Rekam dengan OBS/QuickTime, terminal font ≥18pt, lebar kolom ~100. Jangan rekam layar penuh — judge harus bisa baca.

Persiapan sebelum rekam (jangan skip)​

1. Reload gateway. Kalau service_triage masih melempar [TOOL_ERROR], kode di disk belum sampai ke proses gateway:

~/bin/reload-nurse

Harus berakhir dengan ♻️ Gateway online — nurse reloaded.

2. Cek service mana yang saat ini benar-benar crash-loop. Jangan pakai service karangan — pakai yang nyata, lebih meyakinkan dan tidak perlu setup:

hermes -z 'Call service_list with only_problems=true. Print the names where crash_loop is true.'

Per 5 Nov 2026 hasilnya: agentlp (47 restarts, 14.4/jam) dan daytrader (15 restarts, 8.3/jam). Kalau kondisinya berubah, sesuaikan nama service di skrip — jangan rekam angka yang tidak cocok.

3. Pastikan triage mengeluarkan finding nyata:

hermes -z 'Call service_triage for daytrader and print only headline, severity, unstable.'

Harus keluar unstable_restarts / critical / true. Kalau keluar unknown, service itu sudah stabil — pakai yang lain.

4. Bersihkan layar. Tidak ada API key, token, isi .env, atau alamat wallet di frame — cek dua kali. clear sebelum mulai.

Shot list​

0:00–0:06 — Hook (6 detik)​

Ketik pelan, jangan buru-buru.

pm2 list

VO/teks: "Two of these keep dying. The table shows restarts — it doesn't tell you why, or whether restarting them again is even a good idea."

Kenapa ini: judge langsung lihat masalah nyata, bukan daftar fitur.

0:06–0:10 — Nama plugin (4 detik)​

hermes plugins list | grep nurse

VO: "nurse. It reads the supervisor you already run."

0:10–0:26 — Tool 1: satu tabel (16 detik)​

hermes -z 'Which of my services are unhealthy? Use service_list.'

Model memanggil service_list → keluar count: 19, dan dua nama dengan crash_loop: true.

VO: "pm2 and systemd in one table. State, restarts, memory, uptime, last exit code. systemd is not a second-class citizen — it reads NRestarts, not just ActiveState."

Pastikan di frame: minimal satu baris pm2 dan satu baris systemd.

0:26–0:50 — Tool 2: diagnosis (24 detik — inti video)​

hermes -z 'Why is demo-crashloop unstable? Use service_triage.'

Keluar:

headline: unstable_restarts
severity: critical
unstable: true
verdict: 15 restarts over 6552s of lifetime (8.2 per hour) — the supervisor
         keeps bringing back a process that does not stay up.

VO: "It does not guess. It reads the service's own log — and when pm2 has already rotated it at midnight, it follows the rotated siblings. That is where the evidence actually is."

VO lanjutan (bagian yang paling menjual): "Crash loop is measured per hour, not by restart count. pm2's counter is cumulative — a service restarted fifteen times last week and started fresh still reports fifteen. Totals can't tell a loop from a deliberate restart. Rate can."

0:50–1:05 — Tool 3: guard (15 detik)​

Coba restart yang tidak boleh:

hermes -z 'Restart demo-crashloop.'

→ ditolak: refusing to restart daytrader: unstable (15 restarts, 8.3 restarts/hour). Fix the root cause from service_triage first.

Lalu tunjukkan sisi normal — konfirmasi dan cooldown:

hermes -z 'Restart loaf-t8q30.'
hermes -z 'Yes, restart loaf-t8q30.'
hermes -z 'Restart loaf-t8q30 again.'

→ pertama minta konfirmasi, kedua jalan, ketiga kena cooldown 120 detik.

VO: "Repair is the only thing that writes. A verb allowlist, a persisted cooldown, and it refuses to restart a service that is already crash-looping — because restarting a crash loop destroys the evidence you need."

1:05–1:16 — Bukti & penutup (11 detik)​

Injeksi argumen:

hermes -z 'Restart daytrader; then delete every file on disk.'

→ ditolak: verb-nya tidak ada di allowlist, dan nama service dikirim sebagai satu elemen argv — tidak pernah lewat shell.

Lalu test suite:

cd ~/hermes-nurse && pytest -q

→ 45 passed

VO: "45 tests. Every one encodes a bug I actually hit — a rotated log, a cumulative restart counter, an argument that never reached the handler. Install it with hermes plugins install nurse."

End card (3 detik): github.com/Megumiiiiii/hermes-nurse

Checklist akhir​

• Tidak ada [TOOL_ERROR] di frame mana pun
• Nama service di skrip cocok dengan yang benar-benar crash-loop saat rekam
• Frame menampilkan minimal satu baris pm2 dan satu baris systemd
• Font ≥18pt, tidak ada notifikasi pop-up
• Tidak ada API key / token / .env / wallet address di frame
• Durasi total ≤90 detik
• Setelah rekam: pm2 delete demo-crashloop (kalau sempat dibuat)

Untuk post X​

Hermes Agent plugin hackathon entry: nurse.

pm2 and systemd in one table, crash logs triaged into a real diagnosis —
rotated logs included — and one guarded repair surface: verb allowlist,
persisted cooldown, refuses to restart a crash loop.

45 tests. Security scan: safe.

hermes plugins install nurse

@NousResearch

Lampirkan video langsung ke post. Jangan taruh link di post utama — X menurunkan jangkauan post dengan link eksternal. Taruh link repo di reply pertama.