table-schema

WILLOSCAR's avatarfrom WILLOSCAR

Define evidence-first table schemas for a survey (NO PROSE): what each table must answer, columns, and which evidence-pack fields are required to fill it.**Trigger**: table schema, schema-first tables, table design, 表格 schema, 先 schema 后填充.**Use when**: you want tables to be verifiable and non-placeholder before LaTeX (typically Stage C4 after evidence packs exist).**Skip if**: `outline/table_schema.md` already exists and is refined (no placeholders; >=2 tables defined).**Network**: none.**Guardrail**: NO PROSE; do not invent facts; schema must be checkable and map each column to an evidence source.

⭐1stars🔀0forks📁View on GitHub🕐Updated Jan 10, 2026
[Academic Research]

When & Why to Use This Skill

This Claude skill specializes in designing evidence-first table schemas for research surveys and technical documents. By prioritizing information architecture over prose, it ensures that every table column is strictly mapped to verifiable evidence sources, eliminating placeholders and 'ugly tables' while maintaining high data integrity for LaTeX or formal report generation.

Use Cases

  • Case 1: Designing structured comparison tables for academic literature reviews to ensure every data point is grounded in specific evidence-pack fields.
  • Case 2: Creating verifiable data frameworks for technical surveys where information architecture must be finalized before the drafting stage to prevent factual errors.
  • Case 3: Standardizing complex data presentation in research papers by defining clear row units and column styles that map directly to subsection briefs and evidence drafts.
nametable-schema
description|
Define evidence-first table schemas for a survey (NO PROSE)what each table must answer, columns, and which evidence-pack fields are required to fill it.
**Trigger**table schema, schema-first tables, table design, 表格 schema, 先 schema 后填充.
**Use when**you want tables to be verifiable and non-placeholder before LaTeX (typically Stage C4 after evidence packs exist).
**Skip if**`outline/table_schema.md` already exists and is refined (no placeholders; >=2 tables defined).
**Network**none.
**Guardrail**NO PROSE; do not invent facts; schema must be checkable and map each column to an evidence source.

Table Schema (Evidence-first, NO PROSE)

Purpose: prevent “ugly tables” by fixing the information architecture first (questions + columns + evidence sources), before any filling.

Inputs

  • outline/outline.yml
  • outline/subsection_briefs.jsonl
  • outline/evidence_drafts.jsonl
  • Optional: GOAL.md

Outputs

  • outline/table_schema.md

Non-negotiables

  • NO PROSE: schemas are bullets-only.
  • No placeholders: no TODO, (placeholder), <!-- SCAFFOLD -->, or ellipsis (…).
  • Each table must define:
    • the question it answers
    • the row unit (subsection / cluster / paper)
    • each column + its allowed content style (short phrases; no long prose cells)
    • evidence mapping: which evidence-pack fields are required for each column

Helper script

  • python .codex/skills/table-schema/scripts/run.py --help
  • python .codex/skills/table-schema/scripts/run.py --workspace <ws>

Script

Quick Start

  • python .codex/skills/table-schema/scripts/run.py --help
  • python .codex/skills/table-schema/scripts/run.py --workspace <ws>

All Options

  • See --help.
  • Reads: outline/outline.yml, outline/subsection_briefs.jsonl, outline/evidence_drafts.jsonl.
  • Optional: GOAL.md (for scope + table questions).

Examples

  • Create schema-first table specs:
    • Ensure outline/evidence_drafts.jsonl exists so columns can be tied to evidence fields.
    • Run: python .codex/skills/table-schema/scripts/run.py --workspace workspaces/<ws>

Troubleshooting

Issue: schema columns are not grounded in evidence

Fix:

  • For each column, specify exactly which evidence-pack fields will populate it; remove columns that would require inventing facts.

Issue: schema is too broad (cannot be filled)

Fix:

  • Narrow the table question; prefer two smaller, fillable tables over one “all-in-one” table.