Import Key To PDF

Import Key to PDF

Fills the AnnotsJSON field of a PDF record in <DASHBOARD_URL>/admin/ehealth/medicalpdf/ by analyzing the source PDF with the icd10-panel-fill skill and writing the enriched annotation JSON back to the dashboard.

Invocation

import-key-to-pdf <path.pdf> [<dashboard_url>]

• <path.pdf> — absolute path to the source lab PDF (same file used in W03).
• <dashboard_url> — optional; if omitted, ask the user.

When called from import-lab-recform, <PDF_PATH> and <DASHBOARD_URL> are already set — skip Phase 0 and go straight to Step 1.

Output Language

All user-facing output is in English, regardless of conversation language.

Phase 0 — Validate Inputs (standalone only)

1. Check <path.pdf> was supplied; if missing → ❌ No PDF path supplied. STOP.
2. Validate path exists and ends with .pdf; if invalid → ❌ Invalid path: <path> STOP.
3. If <dashboard_url> not supplied, ask:

   Dashboard base URL? (e.g. https://dev-dashboard.dnainsights.ai)
   

4. Ensure Playwright MCP browser is ready and logged in to <DASHBOARD_URL>/admin/.

Step 1 — Fetch current AnnotsJSON from dashboard

1. Navigate to: <DASHBOARD_URL>/admin/ehealth/medicalpdf/
2. Search for the PDF entry that matches the current recform (search by the flattened filename or Description, e.g. CGX - Amedix <YEAR>).
3. Click on the matching entry to open its change page.
4. Read the current value of the AnnotsJSON field.
   • If the field is empty → report ⚠️ AnnotsJSON is empty. Nothing to fill. and STOP.
5. Save the raw JSON value to /tmp/annots-<recform-slug>.json.
6. Report: ✅ Fetched AnnotsJSON (N annotation objects)

Step 2 — Analyze PDF and build ICD10 panel mapping

Goal: produce { "<uuid>": "diagnosis_icd10codes_panel_<Panel>__<ICD>" } for every annotation whose contents looks like diagnosis_icd10codes__<ICD>.

1. Read the full PDF with Claude Read tool (<PDF_PATH>). Do NOT use pypdf. Confirm all pages captured (panels, ICD-10 code lists, panel headers).

2. Load the AnnotsJSON from /tmp/annots-<recform-slug>.json. Identify annotation objects where contents matches the pattern diagnosis_icd10codes__<ICD> (short form, no panel prefix).

3. For each such annotation:

   • Extract <ICD> from contents.
   • From the PDF content, determine which panel owns that ICD code (the panel whose code list contains <ICD>).
   • Build the full value: diagnosis_icd10codes_panel_<Panel>__<ICD>.
   • Record: { "<uuid>": "diagnosis_icd10codes_panel_<Panel>__<ICD>" }.

4. Save the mapping to /tmp/icd_map.json.

5. If any ICD code cannot be matched to a panel in the PDF:

   • List them as unresolved and ask the user to confirm the panel name before continuing.

6. Show the user a compact confirmation table:

   UUID (first 8 chars)  | contents (before)                    | value (after)
   ─────────────────────────────────────────────────────────────────────────────
   a1b2c3d4              | diagnosis_icd10codes__C50.0           | diagnosis_icd10codes_panel_Breast Cancer__C50.0
   ...
   

   Ask: "Apply this mapping? (yes / adjust)" Do NOT proceed to Step 3 until confirmed.

Step 3 — Apply mapping via icd10-panel-fill

Use the icd10-panel-fill skill inline:

• Mapping file: /tmp/icd_map.json
• Target file: /tmp/annots-<recform-slug>.json

Follow the icd10-panel-fill steps exactly:

1. Dry-run check (confirm UUID matches + count changes).
2. Backup target file as /tmp/annots-<recform-slug>.json.bak.
3. Apply → verify.
4. Report: ✅ icd10-panel-fill: N changed, M skipped

Step 4 — Paste updated AnnotsJSON into dashboard and save

1. Navigate back to the medicalpdf change page for this PDF (<DASHBOARD_URL>/admin/ehealth/medicalpdf/ → search → open entry).
2. Clear the AnnotsJSON field.
3. Paste the full contents of /tmp/annots-<recform-slug>.json into the field.
4. Click Save.
5. Wait for the success confirmation page.
6. Take a screenshot: <recform-slug>/screenshots/<NN>-annots-json-saved.png.
7. Report: ✅ import-key-to-pdf done: AnnotsJSON updated and saved.

Final Report Format (standalone)

PDF:        CGX rec_Amedix-flatten-pdf.pdf
Dashboard:  https://dev-dashboard.dnainsights.ai
Annotations fetched:  42
Panel mappings built: 38  (4 unresolved — listed above)
icd10-panel-fill:     38 changed, 0 skipped
AnnotsJSON saved:     ✅