project_skill

Overview

This skill extracts structured metadata fields from every Note in a Collection, returning a new Collection where each Note contains only the requested fields. It is designed for JSON/dict Notes and preserves nested structure (e.g., metadata.uri). Notes missing any requested field are excluded rather than causing an error.

How this skill works

Provide a Collection target and a list of field paths (supports dot notation like metadata.uri). The skill inspects each Note (must be a dict/JSON object), selects the requested keys, and emits a Collection of Notes each containing only those fields. If a Note lacks any requested field it is filtered out; an empty result means no Notes matched the projection.

When to use it

• You need to extract specific metadata (titles, URIs, years, scores) from search or paper results.
• Preparing a slimmed-down dataset before running joins, aggregations, or downstream transforms.
• Creating a list of URIs to feed into a fetch-text or downloader step.
• Filtering out Notes that don't contain required fields for a subsequent pipeline stage.

Best practices

• Ensure the target is a Collection of dict/JSON Notes; single Note or collections of arrays are unsupported.
• Specify only existing field paths; missing fields cause Notes to be excluded (not an error).
• Use dot notation for nested metadata (metadata.title, metadata.uri) to preserve structure.
• Convert arrays to a Collection with split before projecting elements inside arrays.
• Treat an empty output as a valid 'no matches' result, not a failure.

Example use cases

• Extract metadata.uri from search results to build a list of links for fetch-text.
• Project metadata.title and metadata.year from a papers Collection for metadata-driven filtering.
• Select metadata.source_id and metadata.score to rank or analyze search hits.
• Trim Notes to {text, metadata.uri, char_count} before joining with another dataset.