Works with the AI Agents you already use
Stable-ID LLM Brain Swap
Swap LLM weights behind a stable model ID with automated rollback, resource budgeting, and byte-level verification.
$9.99
Stable-ID LLM Brain Swap
Example session with this skill installed
Production llama.cpp server serves agents through the stable model ID 'prod-agent-v1'. A new GGUF drop just landed: swap the weights under the same ID with zero consumer edits, and arm rollback before you touch anything.
- Read your context and instructions
- Compiled the stable-id llm brain
Snapshot first, then swap the indirection — consumers never change:
Backup — cp -a current weights + serving config into a dated snapshot dir (backups/20261120-swap/), record sha256sum of both GGUF files.
2.
Pre-arm rollback — write the rollback card: restore snapshot, restart service, re-probe health. The swap script runs this automatically on health failure; recovery target under 5 minutes, fully offline.
3.
Repoint, don't edit — the stable ID resolves through the registry/config indirection layer. Swap the symlink/target it dereferences to the new GGUF. Zero consumer or systemd unit edits — that's the point.
4.
Restart in a cron gap — read the watchdog's health-script interval first, time the restart inside a quiet window, keep the expected-model variable intact.
5.
Verify — probe the API for the new model's output signature, confirm the buffered-SSE trap isn't hiding a dead backend, then hold the 48-hour soak before declaring victory.
If health fails, the pre-armed contract restores and restarts on its own — you find out from the breadcrumb notice, not a 2 a.m. page.
Connects securely to your tools. The creator never sees your data.
About this skill
You run a local model behind llama.cpp or a wrapper server. Cron jobs, watchdogs, and agent stacks all point at one model ID. Then a new model drop lands, and swapping it means editing a dozen consumers, praying the watchdog cooperates, and hoping you can roll back at 1 a.m. — while a triage job fires every few minutes and every failed minute is a page.
This skill is the runbook that makes the swap boring. It is distilled from real production operations on self-hosted serving stacks: the exact sequence to swap GGUF weights underneath a stable model ID with zero consumer edits, a sub-five-minute rollback armed before you start, and the failure modes that otherwise cost you a night's sleep. Every procedure is written as tight, imperative steps an agent can execute directly — no tribal-knowledge gaps, no "figure out your own paths" hand-waving, and a rollback card you can type from memory at 2 a.m.
What you get
- The ID-stable swap procedure — a ten-step, copy-ready runbook for replacing weights under a constant model ID on llama.cpp-style stacks, including when to schedule the restart so cron and watchdog cycles never notice.
- The four famous traps, pre-solved — systemd
ProtectHomesilently denying your symlinked weights, dynamic backend worker ports that make you benchmark the wrong model, buffered SSE streams that report zero tokens/sec, and reasoning models whose content field stays empty. Each with the exact detection and fix. - A pre-armed auto-rollback contract — the swap script restores and restarts on health failure by itself, plus a written rollback card you fill in before touching anything. Recovery target: under five minutes, fully offline.
- Watchdog and consumer mapping — how to read health scripts as contracts, keep the expected-model variable intact, and time the swap to a cron gap.
- Operational hygiene — snapshot backups in dated directories, breadcrumb notice files for the next operator, and a 48-hour soak discipline that keeps regressions attributable.
If you swap models more than twice a year, this pays for itself the first time you skip a 2 a.m. manual recovery.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Requires a service restart; not a true blue-green zero-downtime solution. Reasoning field mapping (thinking vs content) must be manually verified per model.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 day ago