healthcare-providers-extract（医疗保健提供者提取物）

Healthcare Providers Extract

Structured practitioner extraction from healthcare practice websites, powered by Nimble's web data APIs.

User request: $ARGUMENTS

Before running any commands, read references/nimble-playbook.md for Claude Code constraints (no shell state, no &/wait, sub-agent permissions, communication style).

Instructions

Step 0: Preflight + WSA Discovery

Run the preflight pattern from references/nimble-playbook.md (5 simultaneous Bash calls: date calc, today, CLI check, profile load, index.md load).

Also simultaneously — run WSA discovery and setup:

• mkdir -p ~/.nimble/memory/{reports,healthcare-providers-extract/checkpoints}
• ls ~/.nimble/memory/healthcare-providers-extract/checkpoints/ 2>/dev/null
• Run Layer 1 (vertical) and Layer 3 (general tools) WSA discovery from references/wsa-reference.md. Layer 2 (session-specific) runs after Step 1 when you know the user's specialty.

Classify discovered agents into phases and validate with nimble agent get per references/wsa-reference.md.

From the preflight results:

• CLI missing or API key unset -> references/profile-and-onboarding.md, stop
• Profile exists -> note it for context. Determine mode using smart date windowing from references/nimble-playbook.md:

- Full mode: first run OR last run > 14 days ago - Quick refresh: last run < 14 days ago (re-extract only new/changed pages) - Same-day repeat: if last_runs.healthcare-providers-extract is today, check for existing report at ~/.nimble/memory/reports/healthcare-providers-extract-*[today].md. If found, ask: "Already ran today. Run again for fresh data?"

• No profile -> that's fine. This skill doesn't require onboarding. Proceed to Step 1.

Step 1: Parse Input & Starting Questions

Parse $ARGUMENTS for input type using the Input Parsing Pattern from references/nimble-playbook.md. Key routing:

• URLs detected -> proceed to Step 3
• Specialty + location (no URLs) -> proceed to Step 2 (practice discovery)
• Unclear -> ask (counts as 1 of max 2 prompts)

If input is clear, confirm and ask one shaping question (plain text, not AskUserQuestion):

"Extracting providers from N practice sites. Quick questions: 1. Healthcare vertical? (ophthalmology, dental, dermatology, general, or other) 2. Quick scan (names + credentials only) or full extraction (all 5 fields)?"

If input is ambiguous, use AskUserQuestion (counts as 1 of max 2 prompts):

What practice sites should I extract providers from? - Paste URLs directly (one per line) - Provide a CSV file path or Google Sheet URL with practice URLs - Or describe what you're looking for (e.g., "ophthalmologists in Austin, TX") and I'll find practices first

Skip questions the user already answered in their initial message.

Step 2: Practice Discovery (Optional)

Only if the user provided a specialty + location instead of URLs.

Two input paths into discovery:

Path A — Fresh discovery. User gave specialty + location. Run Layer 2 WSA discovery for session-specific agents:

nimble agent list --limit 50 --search "[specialty]"
nimble agent list --limit 50 --search "[directory-user-mentioned]"

See references/wsa-reference.md for the full discovery strategy, agent evaluation criteria, and healthcare discovery prioritization.

Run all discovery-phase agents simultaneously. Validate params with nimble agent get first.

Path B — Market-finder handoff. User ran market-finder first and wants to extract providers from those results. Read the market-finder output:

cat ~/.nimble/memory/market-finder/{slug}/entities.json 2>/dev/null

Extract practice records. Note: Google Maps results contain place_url (a Maps link) but not the practice's actual website URL. Proceed to Step 2b to resolve real website URLs before site mapping.

After either path: Deduplicate by domain. Present discovered practices:

"Found N practices for [specialty] in [location] across [M] data sources. Proceeding to extract providers from these sites..."

Fallback — if no discovery WSAs were found, or results are sparse (< 3):

nimble search --query "[specialty] in [location]" --max-results 20 --search-depth lite

Step 2b: Resolve Practice Website URLs

Discovery sources (Google Maps, Yelp, BBB) return listing URLs, not practice website URLs. Before site mapping, resolve the actual website for each practice:

1. Check structured data first — Google Maps results often include a website field in the structured output. Use it if present.
2. Extract from listing page — if no website field, extract the Maps listing to find the practice website link: nimble extract --url "[maps-listing-url]" --format markdown
3. Search fallback — if extraction fails: nimble search --query "[practice-name] [city] official website" --max-results 3 --search-depth lite

Skip practices where no website URL can be resolved — note them in the "Data Quality Summary" output section.

Step 3: Site Mapping

Follow the Site Mapping Pattern from references/nimble-playbook.md for each practice URL. Skill-specific settings:

• Keyword weight table: references/provider-extraction-patterns.md
• Page cap: 15 per site
• Fallback query: site:[domain] doctors OR providers OR team

For 6+ practices, use sub-agents (see Sub-Agent Strategy below).

Save checkpoint: ~/.nimble/memory/healthcare-providers-extract/checkpoints/{slug}/mapping.json

Step 4: Page Extraction

WSA shortcuts first: If WSA discovery found agents that extract provider data from healthcare directories, use those for matching practices — structured WSA output is higher quality than parsed markdown.

For all other practices, follow the Page Extraction with Retry pattern from references/nimble-playbook.md. Scale using the Scaled Execution pattern from the same reference.

Save checkpoint: ~/.nimble/memory/healthcare-providers-extract/checkpoints/{slug}/extraction.json

Step 5: Structured Parsing

Parse extracted markdown to identify providers and their fields. Read references/provider-extraction-patterns.md for the 5 core fields, credential regex patterns, and specialty keywords.

For each extracted page:

1. Scan for provider name patterns (Dr. prefix, heading patterns, bold text near credential suffixes)
2. Match credentials using the regex patterns from references/provider-extraction-patterns.md
3. Match specialty using keywords for the detected healthcare vertical
4. Extract contact info (phone regex, appointment URLs, email)
5. Extract education/training mentions

Build structured records:

{
  "name": "Dr. Jane Smith",
  "credentials": "MD, FACS",
  "specialty": "Retinal Surgery",
  "contact": {"phone": "(555) 123-4567", "scheduling_url": "..."},
  "education": "Fellowship: Bascom Palmer Eye Institute",
  "source_url": "https://practice.com/our-doctors",
  "practice_name": "Shore Center for Eye Care",
  "practice_url": "https://practice.com",
  "confidence": "High"
}

Step 6: Deduplication & Confidence Scoring

Follow the Entity Deduplication and Entity Confidence Scoring patterns from references/nimble-playbook.md. Skill-specific dedup rules and the 5-field confidence criteria are in references/provider-extraction-patterns.md.

Step 7: Output

Present results grouped by practice, sorted by confidence within each practice.

# Provider Extraction: [N] Providers from [M] Practices
*[Date] | [H] High, [M] Medium, [L] Low confidence*

## TL;DR
Extracted [N] providers from [M] practice websites. [H] with complete profiles,
[L] with partial data. [Key finding: e.g., "12 of 15 providers are board-certified"].

## [Practice Name] ([domain])

| # | Name | Credentials | Specialty | Contact | Education | Confidence |
|---|------|------------|-----------|---------|-----------|------------|
| 1 | Dr. Jane Smith | MD, FACS | Retinal Surgery | (555) 123-4567 | Fellowship: Bascom Palmer | High |
| 2 | Dr. John Doe | OD | General Ophthalmology | [Book](url) | Residency: Wills Eye | Medium |

[Repeat per practice]

## Data Quality Summary
- **Complete profiles (High):** [N] providers
- **Partial profiles (Medium):** [N] providers — missing: [list common gaps]
- **Minimal profiles (Low):** [N] providers — missing: [list common gaps]

## Sources
[Clickable URL for every page extracted, grouped by practice]

Source links are mandatory. Every provider record must trace back to a source URL.

Step 8: Save to Memory

Make all Write calls simultaneously:

• Report -> ~/.nimble/memory/reports/healthcare-providers-extract-{slug}-{date}.md
• Provider data -> ~/.nimble/memory/healthcare-providers-extract/{slug}/providers.json
• Profile -> update last_runs.healthcare-providers-extract in ~/.nimble/business-profile.json (only if profile exists)
• Follow the wiki update pattern from references/memory-and-distribution.md: update index.md rows for all affected entity files, append a log.md entry for this run.
• Clean up checkpoint (complete run) or keep (partial run)

Step 9: Share & Distribute

Always offer distribution -- do not skip. Follow references/memory-and-distribution.md for connector detection and sharing flow.

Notion: full provider table as a dated subpage. Slack: TL;DR with provider count and confidence breakdown only.

Step 10: Follow-ups

• "Tell me more about Dr. X" -> show full extracted profile
• "Export as CSV" -> generate CSV from providers.json
• "Run on more sites" -> append new practice URLs, extract and merge
• "What's missing?" -> detail the data gaps per provider

Enrichment from discovered WSAs: If Step 0 found enrichment-phase agents (reviews, regulatory, practice details), offer them as immediate follow-ups:

"I also found [N] WSAs that could enrich this data: [brief list]. Want me to run reputation checks or regulatory lookups on these providers/practices?"

See references/wsa-reference.md for enrichment phase mapping and fallback chains.

Sibling skill suggestions:

Next steps: - Run healthcare-providers-enrich to fill data gaps (NPI lookup, board certification verification, additional contact info) - Run healthcare-providers-verify to validate credentials and license status - Run market-finder to discover more practice URLs in this area

Sub-Agent Strategy

For batch extraction (6+ practices), use nimble-researcher agents (agents/nimble-researcher.md) to parallelize site mapping and extraction.

Follow the sub-agent spawning rules from references/nimble-playbook.md (bypassPermissions, batch max 4, explicit Bash instruction, fallback on failure).

Spawn pattern: One agent per practice (or per batch of 3 practices for large jobs). Each agent runs Steps 3-5 for its assigned practices and returns structured provider records.

Single-practice optimization: If only 1-2 practices, run directly from the main context instead of spawning agents.

Fallback: If any agent fails, run those extractions directly from the main context. Never leave gaps in the output.

Error Handling

See references/nimble-playbook.md for the standard error table (missing API key, 429, 401, empty results, extraction garbage). Skill-specific errors:

• No provider pages found: "Couldn't find provider/team pages on [domain]. The site may list staff differently. Want me to try extracting from the homepage or search for this practice on healthcare directories?"
• All extractions returned garbage: "The practice sites appear to be heavily JavaScript-rendered. Retrying with browser rendering..." (auto-retry with --render per the shared pattern)
• Ambiguous practice name: If a URL fails and the user provided a name instead, search for the practice: nimble search --query "[practice name] [location] doctors" --max-results 5 --search-depth lite
• CSV/Sheet parse error: "Couldn't parse the input file. Expected a column with practice URLs. Can you paste the URLs directly instead?"