# Schema Item Grounding

> Skill: schema item grounding

- Skill: `dingxingdi/schema-item-grounding-2` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx skillmds@latest add dingxingdi/schema-item-grounding-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dingxingdi/schema-item-grounding-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: dingxingdi (https://skillmd.com/u/dingxingdi)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dingxingdi/schema-item-grounding-2

---


# Skill: schema item grounding

## 1. Capability Definition & Real Case
* **Professional Definition**: The ability to map a natural-language question onto the correct tables, columns, and join keys inside a known relational schema, even when schemas are large, redundant, or semantically overlapping.
* **Dimension Hierarchy**: Environment Grounding->Retrieval and Alignment->schema item grounding

### Real Case
**[Case 1]**
* **Initial Environment**: The agent is given a relational schema where multiple tables contain superficially similar name fields. The target database has already been retrieved.
* **Real Question**: The first name of students who have both cats and dogs
* **Real Trajectory**: Inspect the student, pet, and ownership-related tables; reject unrelated name-bearing tables; identify the correct entity and relation path; and compose the SQL with the correct grounding.
* **Real Answer**: The correct SQL must use the student name field rather than a semantically tempting but irrelevant person-name field from another schema area.
* **Why this demonstrates the capability**: This case isolates table-and-column grounding from source retrieval. The database is available, but the agent still fails if it maps the question to a semantically similar yet wrong table. It therefore tests exact schema attachment rather than coarse topical retrieval.

---
**[Case 2]**
* **Initial Environment**: A correct table has already been selected, but the join path depends on specific foreign-key style columns. The agent must inspect relation fields instead of guessing them.
* **Real Question**: Return the rows that connect pets to owners through the ownership bridge table.
* **Real Trajectory**: Read the bridge table, identify the correct join columns such as pet identifiers on both sides, and write the join with those exact fields.
* **Real Answer**: A valid answer requires using the correct bridge-key columns rather than omitting or hallucinating the join keys.
* **Why this demonstrates the capability**: This demonstrates that grounding errors occur even after the correct table family is known. The failure is specifically column-level: the SQL breaks if the agent misses the key fields needed to connect the relations. That makes it an archetypal schema item grounding case.

## Pipeline Execution Instructions
To synthesize data for this capability, you must strictly follow a 3-phase pipeline. **Do not hallucinate steps.** Read the corresponding reference file for each phase sequentially:

1. **Phase 1: Environment Exploration**
   Read the exploration guidelines to discover raw knowledge seeds:
   `references/EXPLORATION.md`

2. **Phase 2: Trajectory Selection**
   Once Phase 1 is complete, read the selection criteria to evaluate the trajectory:
   `references/SELECTION.md`

3. **Phase 3: Data Synthesis**
   Once a trajectory passes Phase 2, read the synthesis instructions to generate the final data:
   `references/SYNTHESIS.md`