Skill Agents

Skill Agents

Personal AI-agent skills, workflows, prompts, and setup guides for using AI assistants with local tools.

This repo is the skill and workflow layer. It tells an AI agent how to work: how to plan, debug, review code, organize Obsidian notes, draft Notion payloads, improve prompts, route skills, and run personal daily workflows.

For executable MCP tools, use the companion repo:

ntaffzii/ai-desk-tools

Skill-Agents       = skills, workflows, docs, examples, provider config
ai-desk-tools      = MCP server, Python tools, tests, runtime safety policy
local-llm          = llama.cpp Router, FastAPI Orchestrator, Docker deployment

Recommended Local LLM repository name: ntaffzii/local-llm-orchestrator.

Who This Is For

Use this repo if you want an AI agent to:

• follow reusable SKILL.md instructions
• use workflow playbooks such as ship-feature or daily-personal-agent
• work across Codex, Claude, ChatGPT, Gemini, LM Studio, local LLMs, and MCP clients
• route tasks to the right skill/workflow before loading context
• connect personal notes, Notion, Obsidian, memory, RAG, and MCP tools
• keep personal automation safe with draft-only and read-only defaults

Companion Repo

This repo does not need to contain the MCP server when published separately.

Install or clone the MCP tools from:

ntaffzii/ai-desk-tools

Typical setup:

1. Clone Skill-Agents for skills/workflows/docs.
2. Clone ai-desk-tools for MCP tools.
3. Configure your AI client to run ai-desk-tools/server.py.
4. Tell your local LLM or agent to use Skill-Agents prompts and workflows.

Quick Start

Clone this repo:

git clone https://github.com/ntaffzii/Skill-Agents.git
cd Skill-Agents

Read first:

• [QUICKSTART.md](QUICKSTART.md)
• [docs/COMPLETELOCALAISYSTEM.md](docs/COMPLETELOCALAISYSTEM.md)
• [docs/PROJECTGUIDE.md](docs/PROJECTGUIDE.md)
• [docs/USAGEFORPERSONALANDOTHERS.md](docs/USAGEFORPERSONALANDOTHERS.md)

Then clone MCP tools:

git clone https://github.com/ntaffzii/ai-desk-tools.git
cd ai-desk-tools
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt

Configure LM Studio, Claude Desktop, or another MCP client to run:

ai-desk-tools/server.py

Install Skills

If your agent supports npx skills:

npx skills add ntaffzii/Skill-Agents

Manual install for Codex-style skills:

cp -R skills/* ~/.codex/skills/

Manual install for Claude Code-style skills:

cp -R skills/* ~/.claude/skills/

On Windows PowerShell, adjust the target path:

Copy-Item -Recurse .\skills\* "$env:USERPROFILE\.codex\skills\"

Local LLM Usage

For LM Studio, Ollama, llama.cpp, vLLM, or any local LLM that does not have native skill loading, use:

[examples/local-llm-agent-prompt.md](examples/local-llm-agent-prompt.md)

Recommended flow:

User request
  -> skill-runtime.route_request
  -> prompt-improver if unclear
  -> skill-runtime.build_agent_context
  -> use recommended toolset/tools
  -> answer with verification

The skill-runtime and prompt-improver tools live in:

ntaffzii/ai-desk-tools

LM Studio MCP Setup

1. Clone and install ntaffzii/ai-desk-tools
2. Open LM Studio
3. Go to Program > Install > Edit mcp.json
4. Add a server pointing to ai-desk-tools/server.py

Example config is in:

[examples/lm-studio-mcp-config.json](examples/lm-studio-mcp-config.json)

For model/provider settings, see:

• [config.json](config.json)
• [docs/CONFIGJSONGUIDE.md](docs/CONFIGJSONGUIDE.md)
• [docs/LOCALLLMSETTINGS.md](docs/LOCALLLMSETTINGS.md)
• [docs/PROMPTIMPROVERLOCALMODEL.md](docs/PROMPTIMPROVERLOCALMODEL.md)

Recommended prompt-improver model when available:

LFM2.5-8B-A1B

It is recommended, not required.

Repo Layout

Skill-Agents/
  Skill.md/        portable model-neutral skills
  skills/          local skill folders with SKILL.md
  workflows/       ordered workflow playbooks
  data/            tools/toolsets/workflows registry metadata
  docs/            guides and usage docs
  examples/        prompts and MCP client examples
  scripts/         validators and repo utilities
  config.json      provider/model/preset reference

Main Workflows

• ship-feature - implement and verify a feature end to end
• debug-regression - reproduce, diagnose, fix, and verify a regression
• review-pr - review a diff or PR with findings first
• obsidian-vault-cleanup - organize Obsidian notes safely
• create-new-skill - create a new reusable skill
• build-mcp-tool - design a new MCP tool
• research-report - create sourced research reports
• daily-personal-agent - build a personal daily plan
• personal-knowledge-sync - connect Obsidian, Notion, memory, and RAG

Skill Categories

• engineering/ - diagnose, TDD, code review, architecture review
• research/ - news, GitHub skill research, project discovery
• productivity/ - handoff, prompt improvement, reports, automation, inbox triage
• obsidian/ - Markdown, Canvas, vault workflow, Notion bridge
• misc/ - skill and MCP tool management
• personal/ - private/personal preferences and daily-agent workflow
• in-progress/ - draft skills
• deprecated/ - retired skills with replacement guidance

Portable Skill.md

Skill.md/ contains portable skills that can be copied into other agents:

• universal-agent-operating-system
• model-portability-adapter
• tool-agnostic-mcp-routing
• cross-model-handoff

Use these when a provider does not support native skill folders.

Common Prompts

Daily planning:

Use daily-personal-agent. Build today's plan from my Obsidian notes, Notion context, memory, calendar, inbox, chat, and open issues. Draft replies but do not send anything.

Obsidian to Notion:

Use personal-knowledge-sync. Treat Obsidian as source of truth. Draft Notion payloads only and preserve tags, wikilinks, aliases, and source paths.

Code work:

Use ship-feature. Inspect docs, repo index, tests, CI, and git before editing. Verify with the narrowest test.

Review:

Use review-pr. Inspect the diff and related files. Report findings first with severity, file, and line reference.

Local LLM routing:

Use skill-runtime first. Route this request, improve it only if unclear, then load selected workflow and skills.

Documentation

• [QUICKSTART.md](QUICKSTART.md)
• [docs/PROJECTGUIDE.md](docs/PROJECTGUIDE.md)
• [docs/USAGEFORPERSONALANDOTHERS.md](docs/USAGEFORPERSONALANDOTHERS.md)
• [docs/SKILLRUNTIMEFLOW.md](docs/SKILLRUNTIMEFLOW.md)
• [docs/LOCALLLMSETTINGS.md](docs/LOCALLLMSETTINGS.md)
• [docs/PROMPTIMPROVERLOCALMODEL.md](docs/PROMPTIMPROVERLOCALMODEL.md)
• [docs/CONFIGJSONGUIDE.md](docs/CONFIGJSONGUIDE.md)
• [docs/TOOLSINVENTORY.md](docs/TOOLSINVENTORY.md)
• [docs/MCPAPISERVER.md](docs/MCPAPISERVER.md)
• [docs/WEBCAPTUREPROVIDERS.md](docs/WEBCAPTUREPROVIDERS.md)
• [docs/FINANCEMARKETPROVIDERS.md](docs/FINANCEMARKETPROVIDERS.md)
• [NOTICE.md](NOTICE.md)

Validate

From the combined workspace:

powershell -NoProfile -ExecutionPolicy Bypass -File .\scripts\validate-all.ps1

This validates:

• tool registry metadata
• toolsets
• workflows
• skills/**/SKILL.md
• Skill.md/**/SKILL.md
• MCP tool Python syntax and tests when tools are present

Attribution

This repo is original work inspired by public agent-skill and MCP repositories. It is not an official fork, template, distribution, or endorsement of those projects.

See [NOTICE.md](NOTICE.md) for attribution and license notes.

License

See [LICENSE](LICENSE).

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ntaffzii
• Source: ntaffzii/Skill-Agents
• License: MIT
• Homepage: https://github.com/ntaffzii/ai-desk-tools

Install and usage instructions live in the source repository linked above.