Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add TserenTserenov/FMT-exocortex-template --skill local-llmgit clone --depth 1 https://github.com/TserenTserenov/FMT-exocortex-templateWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/tserentserenov/fmt-exocortex-template/local-llm)<a href="https://agentmods.dev/skills/tserentserenov/fmt-exocortex-template/local-llm"><img src="https://agentmods.dev/badge/skills/tserentserenov/fmt-exocortex-template/local-llm/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/tserentserenov/fmt-exocortex-template/local-llm"><img src="https://agentmods.dev/badge/skills/tserentserenov/fmt-exocortex-template/local-llm.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium analysis-evasion · line 1 Suspicious Unicode normalization or mixed-script contentFix: Review the flagged content for security risks. Ensure no credentials, secrets, or sensitive data are exposed.
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00078 | $0.01674 |
| Opus 5 | $0.00039 | $0.00837 |
| Sonnet 5 | $0.00016 | $0.00335 |
| Haiku 4.5 | $0.00008 | $0.00167 |
Grade A, and why
local-llm scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 7d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 114 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Локальный LLM-стек (local-llm)
Scope: NL-вход к локальному LLM-стеку (установка / запуск / статус / переключение модели) + слой суждения для оценки находок монитора новых моделей. Not in scope: механика (установка, состояния, переключение, архив) — она в детерминированных скриптах (различение «Скрипт ≠ Агент»). Скилл их вызывает, не дублирует логику. Источник истины статусов моделей — каталог +
model-lifecycle.py. JOB стека: приватность (задачи с персональными данными, что нельзя слать в облако) + запасной режим (без сети / кончились токены облака). См.ADR-001-local-llm-stack.md.
Расположение бандла
Управляющие скрипты — самодостаточный бандл (каждый работает из своей директории).
Путь резолвится через env, дефолт — extensions/local-llm/ внутри IWE-установки:
BUNDLE="${IWE_LOCAL_LLM_DIR:-$HOME/IWE/extensions/local-llm}"
- venv стека:
~/.iwe-local-llm/.venv(ставит установщик). - Активная модель + состояние сервера:
~/.iwe-local-llm/(active-model,server.pid,server.log).
When to use
- «поставь / установи локальную модель» → установка под железо (Шаг 1).
- «покажи модели» / «какие модели есть» → список по состояниям (Шаг 2).
- «запусти / останови» / «статус» → управление сервером (Шаг 2).
- «переключи модель» / «переключи на 14B» → смена активной модели (Шаг 3).
- «какие новые модели вышли» / «проверь новые модели» → мониторинг + суждение об adopt (Шаг 4).
Algorithm
Шаг 1 — установка (механика → скрипт)
bash "$BUNDLE/install-local-llm.sh" # рекомендованная модель под железо
bash "$BUNDLE/install-local-llm.sh" --max # самая тяжёлая влезающая
Установщик идемпотентен: детект железа → выбор модели из каталога под память → venv + mlx-lm → скачивание → smoke. Показать пользователю итог (модель, что сервер поднялся).
Шаг 2 — запуск / статус / список (механика → скрипт)
bash "$BUNDLE/iwe-local-llm.sh" start # поднять сервер (OpenAI-совместимый, localhost:8080)
bash "$BUNDLE/iwe-local-llm.sh" status # активная модель + состояние сервера
bash "$BUNDLE/iwe-local-llm.sh" models # все модели по четырём состояниям + факт скачивания
bash "$BUNDLE/iwe-local-llm.sh" stop # остановить сервер
bash "$BUNDLE/iwe-local-llm.sh" test # проверка совместимого интерфейса (SHIM_OK)
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 7d ago Changed · +1 lines 3c3e64045c40
- 12d ago First seen · 113 lines · 78 tokens per session scan A 4e107b8d611d
local-llm is a skill published in the GitHub repository TserenTserenov/FMT-exocortex-template (53 stars, last pushed today), licensed MIT. It adds 78 tokens to every session and 1,674 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
leakage-guard
Use whenever building features for time-series or any temporal dataset. Enforces strict temporal integrity: no future data in features, no post-event information, correct CV strategy.
test-driven-data-science
Use before any training step to enforce three-layer data validation: Physical (schema), Logical (business rules), Statistical (distribution drift). Blocks training if any CRITICAL assertion fails.
executing-plans
Use when executing a written analysis plan task by task. Manages task state, enforces two-stage review (statistical first, then code quality), and gates manifest updates behind completed reviews.
feature-engineering
Use when creating, transforming, or selecting features. Enforces leakage-free, reproducible feature pipelines.
model-evaluation
Use when evaluating a trained model. Enforces one-time test set evaluation, statistical significance testing, and calibration checks.
model-selection
Use when choosing which model(s) to train. Enforces baseline comparison before hyperparameter tuning, and correct metric selection for the task type.