Belajar Hermes Agen · Bagian 55–58

10/15

Bagian 55–58

Bagian 55–58 dari 101 bagian · Belajar 2

55.Tahapan Tingkat Kemampuan

Level 0 — Pengguna

Bisa:

install, chat

Level 1 — operator

Bisa:

config, model, sessions, tools

Level 2 — pengguna tingkat lanjut

Bisa:

skills, memory, cron, browser, MCP

Level 3 — Agen pembangun

Bisa:

delegation, multi-agen, bots, hooks, plugins

Level 4 — Pengembang Hermes

Bisa:

read architecture, add tools, build plugin, extend gateway, extend provider

Level 5 — Agen sistem insinyur

Bisa mendesain:

LLM
+
context
+
retrieval
+
tools
+
siklus agen
+
memory
+
security
+
evals
+
observabilitas
+
operations

Diagram — Tahapan Tingkat Kemampuan

56.Latihan Laboratorium

lab 1 — asisten Linux

Prompt:

Periksa disk, memori, CPU, dan proses teratas. Jangan mengubah apa pun. Kembalikan bukti dan temuan secara terpisah.

lab 2 — analis log

Analisis sample.log. Group errors by root symptom. Do not assume root cause without evidence.

lab 3 — Skill

Buat skill linux-health-check.

lab 4 — Memory

Ajarkan fakta non-sensitive yang relevan terhadap alur kerja.

lab 5 — MCP

Connect satu hanya-baca data sumber.

lab 6 — Delegation

Parallel riset dari tiga sumber.

lab 7 — Cron

menjalankan daily report.

lab 8 — Gateway

Deliver report ke platform pesan.

lab 9 — Plugin

membangun satu internal diagnostic alat.

lab 10 — keamanan

membuat attack cases for injeksi prompt and agensi berlebihan.

57.Proyek Puncak

Hermes Agent Operasi Perusahaan

Requirements

Agen harus:

Chat via CLI
Keep sessions
Keep curated memory
Use skills
Search web
Read log
Run safe terminal checks
Query controlled enterprise API through MCP
Delegate investigation
Run scheduled health check
Send report to messaging channel
Log failures
Require approval for high-risiko action
Produce evidence-backed findings

arsitektur

                         USER
                          |
                 Telegram / Teams / CLI
                          |
                          v
                  Hermes Gateway/Core
                          |
        +-----------------+-----------------+
        |                 |                 |
      Model             Context          Security
        |                 |                 |
        |          +------+------+          |
        |          |             |          |
        |       Memory         Skills       |
        |          |             |          |
        +----------+------+------+----------+
                          |
                    Agen Loop
                          |
          +---------------+---------------+
          |               |               |
       Native Tools      MCP         Delegation
          |               |               |
      terminal        DB/API/etc.     Subagens
          |
      sandbox
                          |
                       Cron
                          |
                      Reports

Diagram — Proyek Puncak

58.Daftar Periksa Produksi

Model

[ ] model identified
[ ] penyedia identified
[ ] fallback diuji
[ ] rate limits known
[ ] cost measured

Context

[ ] sistem instructions ditinjau
[ ] skill ditinjau
[ ] memori ditinjau
[ ] konteks injection diuji
[ ] long session behavior diuji

Tools

[ ] setiap alat has clear tujuan
[ ] permissions dokumened
[ ] destructive alat restricted
[ ] alat errors handled
[ ] keluaran size controlled

MCP

[ ] servers enumerated
[ ] kredensial isolated
[ ] hanya diperlukan alat exposed
[ ] kesehatan server diuji
[ ] untrusted hasils treated as data

keamanan

[ ] authorization configured
[ ] command approval enabled di mana needed
[ ] sandboxing evaluated
[ ] secret handling ditinjau
[ ] berkas access constrained
[ ] injeksi prompt diuji

Automation

[ ] cron jobs dokumened
[ ] schedules ditinjau
[ ] duplicate execution behavior diuji
[ ] delivery target verified
[ ] failure notification defined

Reliability

[ ] retries understood
[ ] timeouts configured
[ ] long tugas behavior diuji
[ ] session recovery diuji
[ ] backup/restore strategy defined