Korean Analyst Report Fetch + Extract

Given a kr-analyst-report-discover index JSON, download each report's PDF and convert it to plain text. Feeds kr-analyst-report-insight, which writes the final digest.

When To Use

• After kr-analyst-report-discover has produced an index JSON for a ticker.
• When an existing discovered index has new rows you want to refresh into the extracted cache.

Do not use this skill to:

• Discover new reports — that's kr-analyst-report-discover.
• Produce a thesis or digest — that's kr-analyst-report-insight.
• Extract text from image-only PDFs — pypdf only returns selectable text. Image-heavy reports will come back with textLength: 0.

Dependencies

• Node.js (stdlib only).
• Python 3 with pypdf. Install once:

  python3 -m pip install pypdf

• kr-web-browse (for downloadFile).
• Reuses the PDF text extractor from skills/kr-stock-dart-analysis/scripts/extract-pdf-text.py via child_process — do not duplicate the script.

Pipeline

1. Read the input index JSON from --input.
2. Plan downloads: for each report in the index, skip if requiresAuth: true or pdfUrl is null. Compute cache paths:
   • PDF: <cache-dir>/pdfs/<ticker>/<reportId>.pdf
   • TXT: <cache-dir>/pdfs/<ticker>/<reportId>.txt
3. Respect cache: if both files exist and --no-cache was not passed, skip the download + extraction for that report.
4. Download each PDF via kr-web-browse.downloadFile. On HTTP 4xx or content-type mismatch (HTML instead of PDF), mark extractionOk: false with an error string and move on.
5. Extract plain text by invoking the Python script:

   python3 skills/kr-stock-dart-analysis/scripts/extract-pdf-text.py \
       --input <pdf-path> --output <txt-path>

6. Measure: capture textLength (bytes) and pages (from the raw extracted string — the Python script already paginates with \n\n).
7. Cap runtime if --max-reports N was passed.
8. Write a summary JSON to the --output path.

Read references/workflow.md for failure-mode details and the cache layout. Read references/output-format.md for the summary JSON schema.

Operating Rules

1. Never rewrite or compress the extracted text. The .txt file is a verbatim rendering of the PDF's selectable text. Downstream skills quote from it directly.
2. Cache is immutable — same reportId should never trigger a re-fetch unless --no-cache is set. Analyst reports are dated artifacts; their content does not change after publication.
3. Respect the 1-second inter-request delay between PDF downloads (enforced manually here since downloadFile bypasses the browse binary).
4. Never mutate the input index JSON. Write a new file to --output.
5. Fail per report, not per run. A single broken PDF should not stop the rest of the batch. Record the error in reports[i].error and continue.
6. Do not run OCR. If textLength === 0, flag the report in the output and let kr-analyst-report-insight skip its bullet extraction.

Scripts

fetch-reports.js

node skills/kr-analyst-report-fetch/scripts/fetch-reports.js \
  --input .tmp/analyst-report-cache/index/066970/2026-04-18.json \
  --output .tmp/analyst-report-cache/extracted/066970/2026-04-18.json

Optional flags: --max-reports N (default 0 = all), --cache-dir PATH, --no-cache, --verbose, --python3 <bin> (override the Python executable).