Ollama Lifecycle Manager
Security checks across malware telemetry and agentic risk
Overview
This skill is a disclosed Ollama model-management checklist with local commands that fit its purpose and are gated by user confirmation for risky actions.
Install this if you want help managing local Ollama models. Before approving any `ollama rm`, large `ollama pull`, Modelfile creation, or shell profile edit, review the model list, project reference scan, recovery path, and exact command to be run.
Publisher note
这个版本基于原始 Ollama lifecycle 管理思路进行了重写和增强，重点不是堆命令，而是把模型管理变成一套可复盘、可维护、低风险的操作流程。 很多 Ollama 用户都会遇到同一个问题：模型越拉越多，别名越来越乱，脚本里到底引用了哪个模型也记不清，想清理又怕误删正在被工作流使用的模型。这个 Skill 正是为了解决这个问题。 新版强化了几个关键能力： 删除模型前先检查用途、引用和替代方案； 用模型台账记录每个模型的角色、状态和可删除性； 用脚本引用扫描发现哪些模型仍在被项目调用； Benchmark 不只看“总耗时”，而是建议使用 Ollama API 的统计字段； no-think 部分明确区分“提示词要求不展示思考”和 API 层面的 think: false； ModelScope / 镜像回退不只强调下载速度，也提醒来源、tag、量化格式和模板差异风险。 它适合那些已经不满足于“随便拉几个模型试试”的用户，而是希望真正管理好自己的本地 AI 工作台、长期工作流和多模型体系。This version rewrites and extends the original Ollama lifecycle management idea. Instead of being just a list of useful commands, it turns local model management into a safer, more repeatable, and more maintainable workflow. Many Ollama users eventually face the same problem: too many models, unclear aliases, forgotten script dependencies, and the constant risk of deleting something that is still used by a local workflow. This skill is designed to solve exactly that problem. This improved version adds several practical safeguards: Check usage, references, and replacements before deleting a model. Maintain a model inventory with each model’s role, status, and cleanup decision. Scan project files to discover which models are still being referenced. Benchmark with Ollama API metrics instead of relying only on wall-clock time. Clarify the difference between prompt-level “do not show reasoning” and API-level think: false. Treat ModelScope and mirror fallback as useful options while still checking source, tag, quantization, template, and compatibility risks. It is built for users who have moved beyond casual model testing and now want to keep a serious local AI workspace stable, understandable, and safe to evolve.
SkillSpector
By NVIDIA
Vulnerability Patterns
- Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
- Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
- Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
- Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
- Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
VirusTotal
65/65 vendors flagged this skill as clean.