What problem does it solve? CLI-based automation breaks silently when a Claude account hits its weekly or usage limit, and naive retry logic wastes time re-running on transient network or auth errors. This Skill implements a disciplined fallback that switches to Codex only when a genuine usage-limit message is detected. ## Core Features & Use Cases - Conditional fallback: Runs Claude Code first, then re-executes the same prompt with Codex CLI only when output matches real usage-limit phrases like "weekly limit" or "usage limit". - Secure prompt handling: Stores prompts in a mode-600 mktemp file deleted on exit, and reads the OAuth token from ~/.config/llm/claude-oauth-token without logging or transmitting it. - Failure transparency: If Codex also fails, both errors are recorded together instead of emitting a fake successful result. - Use Case: A cron-driven content pipeline calls llm-with-fallback.sh nightly; when the Claude weekly limit resets mid-week, jobs automatically complete via Codex without manual intervention. ## Quick Start Pipe your prompt into scripts/llm-with-fallback.sh with optional max-turns and model arguments to run Claude first with automatic Codex fallback on usage limits.
