llmfiles

This repo is to LLM coding assistants what dotfiles are to your shell, a single place to manage your configuration, symlinked wherever tools expect it. The name comes from there.

Manages skills, commands, and agent instructions in one place to work with Claude Code, Copilot, OpenCode, and Antigravity.

Structure

llmfiles/
├── skills/
│   └── <skill-name>/
│       └── SKILL.md
├── commands/
│   └── <command-name>.md
├── scripts/
│   └── <script-name>.py
├── shared/
│   ├── AGENTS.md
│   ├── glab-mr-context.md
│   └── review-lenses.md
├── CLAUDE.md
├── README.md
└── TODO.md

TODO.md holds work that is scoped but not built yet, so a decision made once does not have to be re-derived later.

File Formats

Skills (skills/<name>/SKILL.md)

Reusable knowledge modules that AI assistants auto-activate based on context.

---
name: skill-name
description: When to use this skill (max 1024 chars)
---

Instructions and knowledge go here.

Commands (commands/<name>.md)

Slash commands triggered with /command-name.

---
description: Brief description
allowed-tools: Read, Grep, Glob
---

Prompt content here. Use $ARGUMENTS for user input.

Scripts (scripts/<name>.py)

Deterministic data gathering a command would otherwise re-derive from scratch every run. Emit JSON on stdout so the command can reason over the result instead of over shell quoting.

Commands call them by absolute path (python3 ~/llmfiles/scripts/<name>.py), so no symlink is needed.

• daily-work-recap.py, one day of git commits, merge requests, GitLab review activity and matching Claude Code transcripts, used by /dailyworkrecap.

AGENTS.md (shared/AGENTS.md)

Universal instructions for AI assistants. Works across VS Code Copilot, Antigravity, and others.

Shared protocols (shared/<name>.md)

Content reused by several commands, extracted so it lives in one place. Commands load it with an explicit read, for example Read ~/llmfiles/shared/review-lenses.md.

Deliberately not skills: a skill is auto-listed with its description and can activate in unrelated sessions, whereas a shared protocol should load only when a command asks for it. Use a skill when the knowledge should surface on its own (glab-cli), a shared protocol when it should not (review-lenses).

Current protocols:

• review-lenses.md, the lens catalog and parallel fan-out contract, used by /glablensedreview, /locallensedreview.
• glab-mr-context.md, GitLab MR setup (batched fetch, read-only MR-head worktree, diff anchor map, prior-pass marker check) plus the goodhabitz lens bindings, used by /glablensedreview and the per-MR agents of /glabsummary.

Installation

Claude Code

# Global instructions
ln -sf ~/llmfiles/shared/AGENTS.md ~/.claude/CLAUDE.md

Option A: Symlink entire directories (recommended)

Easier to maintain: new files automatically appear, deletions propagate cleanly.

# Skills and commands (entire directories)
ln -sf ~/llmfiles/skills ~/.claude/skills
ln -sf ~/llmfiles/commands ~/.claude/commands

Option B: Symlink individual files

Use this if you have existing skills/commands you want to preserve alongside llmfiles.

# Skills (symlink individually)
mkdir -p ~/.claude/skills
for skill in ~/llmfiles/skills/*/; do
  ln -sf "$skill" ~/.claude/skills/
done

# Commands (symlink individually)
mkdir -p ~/.claude/commands
for cmd in ~/llmfiles/commands/*.md; do
  ln -sf "$cmd" ~/.claude/commands/
done

OpenCode

OpenCode stores configuration in ~/.config/opencode.

# Global instructions
ln -sf ~/llmfiles/shared/AGENTS.md ~/.config/opencode/AGENTS.md

Option A: Symlink entire directories (recommended)

# Skills and commands
ln -sf ~/llmfiles/skills ~/.config/opencode/skills
ln -sf ~/llmfiles/commands ~/.config/opencode/commands

Option B: Symlink individual files

Use this if you have existing skills/commands you want to preserve alongside llmfiles.

# Skills (symlink individually)
mkdir -p ~/.config/opencode/skills
for skill in ~/llmfiles/skills/*/; do
  ln -sf "$skill" ~/.config/opencode/skills/
done

# Commands (symlink individually)
mkdir -p ~/.config/opencode/commands
for cmd in ~/llmfiles/commands/*.md; do
  ln -sf "$cmd" ~/.config/opencode/commands/
done

Antigravity

Antigravity stores configuration in ~/.gemini and uses GEMINI.md for instructions.

# Global instructions
ln -sf ~/llmfiles/shared/AGENTS.md ~/.gemini/GEMINI.md

Option A: Symlink entire directories (recommended)

# Skills and commands
ln -sf ~/llmfiles/skills ~/.gemini/antigravity/global_skills
ln -sf ~/llmfiles/commands ~/.gemini/antigravity/workflows

Option B: Symlink individual files

Use this if you have existing skills/commands you want to preserve alongside llmfiles.

# Skills (symlink individually)
mkdir -p ~/.gemini/antigravity/global_skills
for skill in ~/llmfiles/skills/*/; do
  ln -sf "$skill" ~/.gemini/antigravity/global_skills/
done

# Commands (symlink individually)
mkdir -p ~/.gemini/antigravity/workflows
for cmd in ~/llmfiles/commands/*.md; do
  ln -sf "$cmd" ~/.gemini/antigravity/workflows/
done