Portolan Bootstrap

Portolan Bootstrap Skill

Bootstrap a complete geospatial data catalog from any source portolan supports.

Critical Principles

1. NEVER hallucinate metadata. Only use information that exists in the source.
2. When uncertain, checkpoint and ask. Prefer asking too much over assuming.
3. Always dry-run first. Warn about large outputs, slow operations, anything concerning.
4. Failures: warn inline, summarize at end. Don't stop for every failure, but don't hide them.

Phase 1: Discovery

Detect source type

Portolan supports two workflows:

Remote services (use portolan extract):

• WFS: /wfs, /geoserver/, service=WFS
• ArcGIS FeatureServer: /FeatureServer, /MapServer
• ArcGIS ImageServer: /ImageServer

Local files (use portolan init → scan → check → add):

• Directory with Shapefiles, GeoJSON, GeoPackage, GeoTIFF, etc.

For remote services: Dry-run extraction

portolan extract <type> "<URL>" --dry-run

For local files: Initialize and scan

portolan init --auto
portolan scan

⏸️ CHECKPOINT: Discovery Results

Present to user:

• Layer/file count
• Estimated total features (if available)
• Any warnings (large counts, mixed formats, non-cloud-native files)

Ask: "Found X layers/files. This may take Y minutes and produce ~Z MB. Proceed?"

Warn if:

• Layer count > 50 (will take time)
• Any layer has > 100k features (memory concerns)
• Mixed geometry types detected
• Missing CRS information
• Non-cloud-native files that need conversion (GPKG, Shapefile, etc.)
• Files at root level (must be in subdirectories for collections)

Phase 2: Extraction / Conversion

For remote services: Run extraction

portolan extract <type> "<URL>" --output <path>

Monitor progress. Warn inline on any failures but continue.

For local files: Check and convert

# Preview what needs conversion
portolan check --dry-run

# Convert non-cloud-native files (GPKG → GeoParquet, etc.)
portolan check --fix

IMPORTANT: Files must be in subdirectories. Each subdirectory becomes a STAC collection. If files are at root level, move them into appropriately named subdirectories first.

⏸️ CHECKPOINT: Extraction/Conversion Complete

Present:

• Success count vs total
• List any failures with reasons
• Total data size
• Files converted (for local workflow)

Ask: "Extracted/converted X/Y layers. Z failures: [list]. Continue?"

Phase 3: Remote Setup

⏸️ CHECKPOINT: Destination

Ask: "Where should this catalog be published?"

• Source Cooperative (s3://us-west-2.opendata.source.coop/...)
• Other S3 bucket
• GCS bucket
• Azure blob storage

⏸️ CHECKPOINT: License

Ask: "What license applies to this data?"

• Show what's in source metadata (if anything)
• Common options: CC-BY-4.0, CC0, ODbL

⏸️ CHECKPOINT: Contact Info

Ask: "Contact info for this catalog?"

• Name
• Email

Validate credentials

# Set up .env
echo "PORTOLAN_REMOTE=<destination>" > .env

# Dry-run push to validate
portolan push --dry-run

Phase 4: Asset Generation

Check tippecanoe availability

which tippecanoe

⏸️ CHECKPOINT: PMTiles Generation

Ask: "Generate PMTiles for web visualization?"

• Note if tippecanoe is/isn't available
• Warn about time for large datasets

Register files and generate assets

# For remote extractions (files already in collection subdirs):
portolan add . --recursive [--pmtiles]

# For local files (after check --fix):
portolan add <collection-dir>/ [--pmtiles]

⏸️ CHECKPOINT: Assets Generated

Present:

• Collections processed
• Thumbnails generated
• Any failures

Phase 5: Metadata Enrichment

CRITICAL: Only use information from the source. Do NOT invent content.

What enrichment means

✅ DO:

• Fill fields that exist in source but weren't auto-extracted
• Re-read source service metadata (GetCapabilities, service info) to find missed fields
• Fix encoding issues (mojibake, garbled characters)
• Clean up formatting (extra whitespace, inconsistent capitalization)
• Standardize inconsistencies within the source (if source uses "Pergamino" and "Perg." → use whichever is more common in source)

❌ DO NOT:

• Invent titles that aren't in the source
• Translate unless source provides both languages
• Add descriptive context that isn't in the original
• Expand abbreviations using external knowledge
• "Improve" wording beyond what the source says

Infer language

Check layer names and descriptions:

• If consistently one language → use that
• If mixed → checkpoint and ask user

⏸️ CHECKPOINT (if mixed language)

Ask: "Source has mixed languages (Spanish titles, English descriptions). Which should be primary?"

Check source metadata completeness

Compare what portolan extracted vs what's in the source:

1. Re-read source service metadata (GetCapabilities, ArcGIS service info, etc.)
2. For each collection, check:
   • Does source have a title that wasn't extracted? → Use it
   • Does source have a description that wasn't extracted? → Use it
   • Does source have attribution/contact info? → Use it

⏸️ CHECKPOINT: Metadata Gaps

If you find fields in the source that weren't extracted:

Present: "I found these fields in the source metadata that weren't auto-extracted:"

• [Show exact text from source for each field]

Ask: "Should I add these to the catalog? (Showing source text exactly as-is)"

⏸️ CHECKPOINT: Remaining Gaps

For fields that MUST be filled but aren't in source:

Present: "These required fields aren't in the source metadata:"

• Catalog title: [missing]
• Catalog description: [missing]

Ask: "Please provide values for these, or I can leave them as the auto-generated defaults."

Update catalog.json

Only with confirmed/approved content:

catalog["title"] = "<from source or user-provided>"
catalog["description"] = "<from source or user-provided>"

Update collection metadata

For fields from source:

collection["title"] = "<from source>"  # Only if source has it

For .portolan/metadata.yaml (per ADR-0038):

contact:
  name: "<from checkpoint>"
  email: "<from checkpoint>"
license: "<from checkpoint>"
attribution: "<from source if available>"
source_url: "<original service URL>"
keywords: "<from source if available>"
processing_notes: "Extracted from <service type> on <date>"

Phase 6: Metadata Templates & README Generation

Create metadata templates (always recursive)

portolan metadata init --recursive

This creates .portolan/metadata.yaml at catalog and all collection levels.

Generate READMEs (always recursive)

portolan readme --recursive

⏸️ CHECKPOINT: README Review

Show catalog-level README summary.

Ask: "READMEs generated. Here's the catalog README. Look correct?"

Phase 7: STAC-GeoParquet (if >1000 assets)

If the catalog has more than 1000 assets, generate items.parquet for efficient queries:

portolan stac-geoparquet

This creates an items.parquet file in each collection for fast spatial/temporal queries.

Skip this step for smaller catalogs.

Phase 8: Push

Dry-run first

portolan push --dry-run

⏸️ CHECKPOINT: Push Confirmation

Present:

• File count
• Total size
• Destination URL

Ask: "Ready to push X files (Y MB) to Z. Proceed?"

Execute push

portolan push --verbose

Phase 9: Final Summary

Present:

• Total collections published
• Live URL
• Any failures/warnings from entire session
• Suggested next steps (verify in browser, check a sample collection)

Failure Handling