Tlb Natural Text To Table

TLB Natural Text To Table

Use this skill when the input is not yet a TileLineBase table, but it already contains repeated, recognizable records such as tasks, people, events, inventory items, requirements, or status updates.

Completion standard

The task is complete only when all of these are true:

• The output is valid TileLineBase row content built from repeated ## field: value blocks.
• The first field name is consistent across every row. For hierarchical output, use one shared primary field such as entry.
• Skill-added field names are written in English. Values extracted from the source keep the source language.
• Parent-child output uses TLBparent: on child rows only. This helper field is for the Conversion Assistant and should not be treated as a visible business field.
• Default output uses a minimal schema: entry, optional description, and optional TLBparent.
• Do not introduce extra fields unless the user explicitly asks for them.
• Values come from the source text; weak guesses are avoided.
• Noise, commentary, and one-off prose are either omitted or called out before conversion.
• If the source mixes multiple record shapes that should not share one table, split them or stop and ask.

Required output shape

TileLineBase row data uses one H2 block per row:

## entry: Write launch post
description: Needs final screenshot

Rules:

• The first line of each row must start with ## .
• The first field should be one shared primary label for every row. Prefer a general English field such as entry unless the user explicitly asks for another stable field name.
• Follow with flat field: value lines.
• If the source contains parent-child structure, keep the same primary field on both parent and child rows, and add TLBparent: <parent primary value> on child rows only.
• By default, only use these skill-added field names: entry, description, TLBparent.
• Do not translate extracted values. Preserve the source language in row titles and field values.
• Keep one field name spelling across all rows.
• Use an empty value only when the field is truly missing and a shared schema matters.
• Do not add free text between rows.

Read references/examples.md only when you need concrete patterns.

Workflow

1. Identify the row unit. Treat each repeated item as one row. Good units are one bullet block, one agenda item, one person card, one changelog item, or one repeated paragraph pattern.

   If the source is hierarchical, still treat each parent or child item as one row. Do not switch the primary field name between parent and child rows.

2. Pick the primary field. Choose the field that best names the row. Usually this is the subject users scan first.

3. Draft a shared field set. Keep it very small and stable. Default to entry plus description. If hierarchy exists, add TLBparent for child rows only.

4. Extract facts conservatively. Preserve source wording when it carries meaning. Normalize only light formatting such as extra spaces, checkbox markers, and duplicate punctuation.

   Preserve the original language of extracted values. Only the structural field names introduced by the skill should be normalized to English.

5. Fill missing values carefully. If a value is absent, leave it empty instead of inventing one. If many rows are missing the same field, drop that field unless the user clearly needs it.

6. Emit clean TLB blocks. Output only the final Markdown unless the user asked for explanation or review.