Skip to main content
Loom Core docs

Agent Auth Sync

This repo owns the operator-facing launchd workflow for keeping cluster agent credentials aligned with the authenticated files on a Mac workstation.

Canonical flow

  1. Install the launchd job with loom sync agent-tokens install.
  2. That installs com.loom.agent-token-sync on macOS.
  3. The launchd job runs platform/gitops/bin/sync-agent-tokens.
  4. The GitOps helper refreshes k3s/devbox/agent-auth-tokens.yaml from:
    • ~/.codex/auth.json
    • ~/.gemini/oauth_creds.json
    • ~/.gemini/google_accounts.json
  5. Flux applies the resulting agent-auth-tokens secret for cluster agents that use file-backed auth.

Claude path

Claude does not use the file-backed sync above.

  • HUD/devbox launches wire Claude through ANTHROPIC_API_KEY.
  • The cluster source of truth for that key is k3s/devbox/agent-api-keys.yaml.
  • The Claude launcher path is implemented in internal/hud/spawn.go.

That means ~/.claude/auth.json is not part of the supported cluster sync story for loom-core.

Relevant code

  • cmd/loom/cmd_sync_agent_tokens.go
  • internal/hud/spawn.go
  • platform/gitops/bin/sync-agent-tokens
  • platform/gitops/k3s/devbox/agent-auth-tokens.yaml
  • platform/gitops/k3s/devbox/agent-api-keys.yaml
Agent Auth Sync | Loom Core docs