Claude Code Paper Digest

For step-by-step debugging:

sh scripts/run_cli.sh discover --config config/paper_fetch.toml --topic <topic> --days <days> --output /tmp/discover.json
sh scripts/run_cli.sh enrich --config config/paper_fetch.toml --input /tmp/discover.json --output /tmp/enrich.json
sh scripts/run_cli.sh rank --config config/paper_fetch.toml --topic <topic> --input /tmp/enrich.json --limit <limit> --output /tmp/rank.json
sh scripts/run_cli.sh annotate --input /tmp/rank.json --annotations /tmp/annotations.json --output /tmp/annotated.json
sh scripts/run_cli.sh render --target markdown --input /tmp/annotated.json --output <output_path>

Required Behavior

• Keep the output in Chinese unless the user explicitly overrides it.
• Before the first CLI call, resolve Python with sh scripts/resolve_python.sh.
• If a compatible interpreter exists, proceed without asking the user to upgrade Python.
• Only if no compatible interpreter exists, or runtime still fails after using the resolved interpreter, ask whether the user wants help locating Python 3.11+, creating a virtual environment, or adjusting the install flow.
• Before running annotate, create /tmp/annotations.json with the following schema — one entry per paper using the arxiv_id values returned by the previous step:

{
  "papers": [
    {
      "arxiv_id": "<id from rank output>",
      "summary_zh": "<full faithful Chinese translation of the English abstract — not a compressed summary>",
      "positive_take": "<one sentence: the paper's key contribution in positive framing>",
      "critical_take": "<one sentence: a sharp, honest critique compared with prior work>"
    }
  ]
}

• Use output_path as the final artifact path.
• For manual runs, prefer --include-seen so the digest is reproducible.