Dotfiles-nix claude-to-agents

install

source · Clone the upstream repo

git clone https://github.com/not-matthias/dotfiles-nix

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/not-matthias/dotfiles-nix "$T" && mkdir -p ~/.claude/skills && cp -r "$T/modules/home/programs/cli-agents/shared/skills/claude-to-agents" ~/.claude/skills/not-matthias-dotfiles-nix-claude-to-agents && rm -rf "$T"

manifest: modules/home/programs/cli-agents/shared/skills/claude-to-agents/SKILL.md

source content

Claude to Agents

Create relative symlinks between Claude-specific config and the generic

.agents/

AGENTS.md

convention defined by the Agent Skills spec.

The

.agents/skills/

path is the official cross-client interop directory scanned by Claude Code, Cursor, Gemini CLI, Amp, VS Code Copilot, OpenCode, Roo Code, Goose, Junie, Codex, Kiro, and many others. See agentskills.io/clients for the full list.

When to Use

User wants the repo to work with non-Claude agents
User asks to "create AGENTS.md", "set up .agents/", or "normalize config"
User wants cross-client skill sharing

Mapping

Claude-specific	Generic equivalent	Sync?
`CLAUDE.md`	`AGENTS.md`	Yes
`.claude/skills/`	`.agents/skills/`	Yes
`.claude/docs/`	`.agents/docs/`	Yes
`.claude/scripts/`	`.agents/scripts/`	Yes
`.claude/SCRATCHPAD.md`	`.agents/SCRATCHPAD.md`	Yes
`.claude/settings.json`	—	No
`.claude/settings.local.json`	—	No
`.claude/tmp/`	—	No

User-level paths follow the same pattern:

Claude user-level	Generic user-level
`~/.claude/skills/`	`~/.agents/skills/`

Workflow

Step 1: Scan for existing files

From the project root:

ls -la CLAUDE.md AGENTS.md .claude .agents 2>/dev/null

Classify each pair:

Source exists, target missing → create symlink
Both exist, one is already a symlink → already synced, skip
Both exist, neither is a symlink → warn user, ask which to keep
Neither exists → skip

Step 2: Determine source of truth

Whichever file/directory is a real file (not a symlink) is the source of truth. If both are real:

Prefer the one with more content or a more recent mtime
If ambiguous → ask the user

Step 3: Create symlinks

Always use relative symlinks so they work when the repo is cloned elsewhere.

# CLAUDE.md is source of truth
ln -s CLAUDE.md AGENTS.md

# Or AGENTS.md is source of truth
ln -s AGENTS.md CLAUDE.md

# Subdirectories — sync individually, not the whole parent
mkdir -p .agents
ln -s ../.claude/skills .agents/skills
ln -s ../.claude/docs .agents/docs

# Reverse direction
mkdir -p .claude
ln -s ../.agents/skills .claude/skills

Step 4: Handle subdirectories carefully

Sync

.claude/

↔

.agents/

per-subdirectory, not by symlinking the entire directory. Each side may have unique content (e.g.

.claude/settings.json

is Claude-only).

Only sync:

skills/

docs/

scripts/

SCRATCHPAD.md

Never sync:

settings.json

settings.local.json

tmp/

, or any client-specific cache/config files.

Step 5: Update .gitignore if needed

Ensure symlinks are tracked but ephemeral files are not:

# Agent config — track symlinks, ignore ephemeral
.claude/tmp/
.agents/tmp/
.claude/settings.local.json

Step 6: Report

Print a summary:

Normalized agent config:
  CLAUDE.md ← source of truth
  AGENTS.md → CLAUDE.md (symlink created)
  .agents/skills/ → .claude/skills/ (symlink created)
  .agents/docs/ — skipped (doesn't exist on either side)

Important Notes

Always use relative symlinks (not absolute paths)
Never overwrite real files with symlinks without asking
The mapping is bidirectional — works regardless of which convention existed first
Only handles Claude ↔ generic mapping; other client-specific files (.cursorrules, .gemini/, etc.) are out of scope
If both CLAUDE.md and AGENTS.md have real, different content, stop and ask — don't silently pick one