Skip to content

Using-AI_Global_Tools.md

Latest commit

 

History

History
162 lines (98 loc) · 3.54 KB

Using-AI_Global_Tools.md

File metadata and controls

162 lines (98 loc) · 3.54 KB
Raw

Using AI_Global_Tools

Non-repo-specific rule

This setup must work outside any particular repo.

Always treat the absolute vault path below as canonical, even if you are operating from another project, a temporary directory, or a different worktree:

  • /mnt/hyperliquid-data/projects/obisidan/AI_Global_Tools

When asked to use a skill, workflow, prompt, or agent asset, resolve it from this vault first instead of searching the current repo for a local copy.

Resolution order:

  1. tool-specific folders in this vault
  2. shared folders in this vault
  3. repo-local copies only when explicitly requested or when no canonical vault copy exists

What is now global

The following are now sourced from this vault:

  • Codex skills
  • Claude Code skills
  • Claude global system prompt
  • Codex global developer instructions
  • Claude MCP definitions
  • Codex config file (including current MCP entries)
  • shared prompt, agent, skill, and template documentation
  • reusable LLM Wiki schema/templates
  • shared herdr skill available to Claude, Codex, and Pi via vault symlinks

Live bridges on this machine

Claude Code

The claude command now resolves through a local wrapper in:

  • ~/.local/bin/claude

That wrapper automatically:

  • appends claude/prompts/system/claude-global.md
  • adds this vault as an accessible directory
  • loads MCP config from mcp/claude/mcp-config.json
  • uses --strict-mcp-config unless you explicitly override MCP flags yourself

Codex

The codex command now resolves through a local wrapper in:

  • ~/.local/bin/codex

That wrapper automatically:

  • exports AI_GLOBAL_TOOLS_DIR=/mnt/hyperliquid-data/projects/obisidan/AI_Global_Tools
  • injects codex/prompts/system/codex-global.md as developer instructions

Also:

  • ~/.codex/skills -> vault symlink
  • ~/.claude/skills -> vault symlink
  • ~/.codex/config.toml -> vault symlink to mcp/codex/config.toml

Where to edit things

Add or edit Claude skills

Edit under:

  • claude/skills/

Add or edit Codex skills

Edit under:

  • codex/skills/

Add reusable prompts

Edit under:

  • shared/prompts/library/
  • shared/prompts/system/
  • claude/prompts/system/
  • codex/prompts/system/

Add reusable shared skills

Canonical shared skills can live under:

  • shared/skills/

Then runtime-specific skill paths can symlink back to them, as with:

  • shared/skills/herdr/

Use or adapt the LLM Wiki pattern

Start with:

  • LLM-Wiki-Pattern.md
  • shared/templates/llm-wiki/

Current live instance created from this pattern:

  • /mnt/hyperliquid-data/projects/obisidan/Evonome-vault

Add or edit MCP definitions

Edit under:

  • mcp/claude/mcp-config.json
  • mcp/codex/config.toml
  • mcp/bin/

Secrets location

Secrets are intentionally not stored in the vault.

Use:

  • ~/.config/ai-global-tools/mcp-secrets.env

There is also a template here:

  • ~/.config/ai-global-tools/mcp-secrets.env.example

Backups

Pre-migration backups were stored in:

  • ~/.local/share/ai-global-tools-backups/

How to check it later

type -a claude
type -a codex
readlink -f ~/.claude/skills
readlink -f ~/.codex/skills
readlink -f ~/.codex/config.toml

Practical rule

If you want a reusable agent tool to become global, add it to this vault first.

If a user says things like:

  • "use the bug-check skill"
  • "use the Claude pre-pr skill"
  • "run that Codex workflow"

then the agent should first look in:

  • claude/skills/
  • codex/skills/
  • shared/skills/
  • related prompt/docs folders in this same vault

and should not assume a repo-local copy is authoritative.