Automated content curation from Product Hunt and Reddit, powered by Claude Code
The digest system runs daily at 9 PM via systemd timer. An orchestrator script manages Claude Code execution with retry logic and completion detection.
The orchestrator manages Claude Code execution, handling failures gracefully with automatic retries.
Claude Code reads CLAUDE.md and executes a 9-step pipeline to generate the daily digest.
Content is discovered from primary sources only. WebSearch/WebFetch are used for verification, never discovery.
All state is persisted to checkpoint.md after every action. This enables crash recovery and ensures data isn't lost if context runs out.
| Phase | Resume Point | Description |
|---|---|---|
starting |
Step 0 | Initialize and scan previous digests |
source-0-complete |
Step 2 | Product Hunt done, start AI/LLMs |
source-1-complete |
Step 3 | AI/LLMs done, start GenAI |
source-2-complete |
Step 4 | GenAI done, start Apps |
source-3-complete |
Step 5 | All sources done, final selection |
selection-complete |
Step 6 | Selection done, generate HTML |
complete |
Done | All steps finished |
Key files and their purposes in the digest system.
The orchestrator uses a strict two-part check to verify completion.
.orchestrator-ack-{date} to prevent re-runs on the same day.