Georgi Gaydarov

Network engineer building multivendor CLI & AI tools for NetOps. 9,800+ commands for Cisco, Juniper, Arista. Automation, observability, less firefighting.

[Georgi Gaydarov @gesh755]

Hot take: prompting is the least important AI skill. (1/5)

What actually changes results is how you operate the model: like infrastructure, not a chat window.

Here's how I run Claude for engineering work 🧵

[Georgi Gaydarov @gesh755]

Phase 6 of the lab: a detected anomaly now triggers the fix, risk-gated.

LOW: auto-executes
MEDIUM/HIGH: one-click approval
CRITICAL: pages a human, never auto-acts

Changes use confirmed-commit (RFC 6241): no confirm, auto-revert.

https://t.co/6e47RfzdYI #NetDevOps

[Georgi Gaydarov @gesh755]

An AI agent with write access to production network gear and no rollback isn't automation. It's an outage with extra steps.

My rules:
- Read-only by default
- Sanitize before the model sees anything
- Fail closed when policy is unclear
- A human owns the blast radius

[Georgi Gaydarov @gesh755]

Build log: claude-skill-lint v0.5.

A Claude Code skill's description loads into context every turn; the body only when it fires. A bloated description is a permanent tax.

The linter flags that, plus dead links, secrets and stale model IDs. Zero deps.

https://t.co/vzUNwHr4rk

[Georgi Gaydarov @gesh755]

HMAC tokens just landed in the gate.

AEGIS is air-gapped change validation for networks that cannot send topology off-prem.

Intent goes in. A self-hosted LLM may propose config. Everything after that is deterministic:
Batfish → containerlab twin → compliance crosswalk → sealed evidence.

LLM proposes. Pipeline verifies. Human authorizes.

v0.2.0 is on main. Phase 2 closed. Phase 3 authenticity just shipped:
HMAC-bound approval tokens + API-key gate. Random strings no longer pass G2/G3.

25,000 adversarial runs. 0 invariant violations.
2,000 concurrent ops. 0 leaks.
11 frameworks / 30 controls.
Egress: none. Structural, not a policy checkbox.

Lab (built with Grok):
https://t.co/eGe3TdZynb

Source:
https://t.co/VnZt0MTJYj

Docs:
https://t.co/xAzy8tmEas

#NetDevOps #NetworkAutomation #AirGapped

[Georgi Gaydarov @gesh755]

Network engineers: what would you let an AI agent do on your production network today with no human approving it?

Read-only show commands? Drafting tickets? Low-risk changes with auto-rollback? Nothing at all?

Curious where the line is for you. #NetDevOps

[Georgi Gaydarov @gesh755]

netlog-ai 1.2.0 adds a SolarWinds connector.

Orion (parameterized SWQL) or the Observability SaaS API: alerts, syslog and events are normalized to one line format, then ranked like any other source. Alert notes and acks happen only on an explicit call.

https://t.co/9tTujN8VBd

[Georgi Gaydarov @gesh755]

netlog-ai v1.1.0 is out.

Spine-leaf link degrades, NOC gets hundreds of BFD/BGP/VTEP alarms. v1.1 finds the lowest-layer root cause and suppresses the noise.

579 tests · 15 MCP tools
https://t.co/7dovupXgqc

[Georgi Gaydarov @gesh755]

netlog-ai 0.7.0: syslog in, ranked actions out, and now a sanitized incident handoff.

The ticket paste card is built from health score, blast radius and change window, then run through the same secret gate as an LLM prompt. Raw syslog stays out.

https://t.co/7dovupXgqc

[Georgi Gaydarov @gesh755]

https://t.co/7dovupXOfK

[Georgi Gaydarov @gesh755]

The paste that goes into the ticket should not be the raw log.

0.7 of netlog-ai builds the shift brief on the machine. Secrets are stripped first. Nothing is posted for you.

[Elon Musk @elonmusk]

This will be cool

[Jacob Coxon @hilbertspaess]

I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.

[Georgi Gaydarov @gesh755]

v0.5 zero-dependency linter for Claude Code Agent Skills.

90+ checks.
2026 model IDs (Opus 5 / Sonnet 5).
Computer-use safety.
22-skill NetOps pack.

Frontmatter, progressive disclosure, dead refs, stale model IDs — caught before the skill hits a fabric.

Source:
https://t.co/DVzR8wKOAN

Docs:
https://t.co/RZVTSSJrhK

#ClaudeCode #AgentSkills #NetOps

[Mahax @Mahaximus_]

SpaceXAI just dropped the best 1-hour course on Grok Bot agents: from 1 prompt to a full 24/7 system

2:15 - Build your first Grok Bot
6:52 - Give every Bot a role
16:51 - Make Bots work 24/7
31:50 - Run agents in parallel
52:18 - full autonomous system

Free, the best thing on Grok Bot agents I've come across.

Watch it, then read the full Grok Bot breakdown below

[Georgi Gaydarov @gesh755]

https://t.co/cNAUAJHZYQ

[Georgi Gaydarov @gesh755]

https://t.co/cNAUAJHZYQ

[Georgi Gaydarov @gesh755]

70,000+ CLI commands.
17 vendors.
One HTML file.
Zero dependencies.

Concept-aligned N-vendor compare.
NETCONF / Ansible drawers.
CLI Studio adds intent, migrate, recipes, EOS/FRR/VyOS parsers, and a 10-node FRR lab map.

Cisco · Juniper · Arista · Nokia · Fortinet · Palo Alto · FRR · VyOS · Huawei · Aruba · Mikrotik · Extreme · NVIDIA · SONiC — plus PowerShell / Linux / Wireshark.

Bookmark it. Search it. Stop grepping vendor PDFs.

Live:
https://t.co/1vnXdBUJ51

Source:
https://t.co/nJ62ffhrDJ

#NetworkEngineering #NetDevOps #CLI

[Georgi Gaydarov @gesh755]

@elonmusk true

[Elon Musk @elonmusk]

𝕏 is where almost all AI discourse takes place

[Georgi Gaydarov @gesh755]

70,000+ CLI commands.
17 vendors.
One HTML file.
Zero dependencies.

Concept-aligned N-vendor compare.
NETCONF / Ansible drawers.
CLI Studio adds intent, migrate, recipes, EOS/FRR/VyOS parsers, and a 10-node FRR lab map.

Cisco · Juniper · Arista · Nokia · Fortinet · Palo Alto · FRR · VyOS · Huawei · Aruba · Mikrotik · Extreme · NVIDIA · SONiC — plus PowerShell / Linux / Wireshark.

Bookmark it. Search it. Stop grepping vendor PDFs.

Live:
https://t.co/1vnXdBUJ51

Source:
https://t.co/nJ62ffhrDJ

#NetworkEngineering #NetDevOps #CLI