# Multi Turn Instruction Following

> Skill: multi-turn-instruction-following

- Skill: `dingxingdi/multi-turn-instruction-following-2` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx skillmds@latest add dingxingdi/multi-turn-instruction-following-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dingxingdi/multi-turn-instruction-following-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/multi-turn-instruction-following-2

---


# Skill: multi-turn-instruction-following
## 1. Capability Definition & Real Case
* **Professional Definition**: The ability to preserve, update, and correctly execute user instructions across consecutive dialogue turns, especially when later turns transform or constrain an earlier response rather than replacing it entirely.
* **Dimension Hierarchy**: Conversational Effectiveness->Open-ended Interaction Quality->multi-turn-instruction-following
### Real Case

**[Case 1]**
* **Initial Environment**: A text-only chat session has already produced a detailed travel blog post about Hawaii. No external tools are needed, but the full earlier answer remains in the dialogue history and must be treated as binding context.
* **Real Question**: Rewrite your previous response. Start every sentence with the letter A.
* **Real Trajectory**: 1. Retrieve the immediately previous answer from dialogue history. 2. Preserve its high-level content units instead of generating a new topic. 3. Rewrite each sentence so the constraint applies globally rather than sporadically. 4. Verify that every sentence begins with the required letter before sending the response.
* **Real Answer**: A travel-oriented rewrite that preserves the original Hawaii content while ensuring every sentence begins with “A.”
* **Why this demonstrates the capability**: The user is not asking for a fresh standalone answer; the turn modifies an earlier output. Success requires carrying forward earlier semantic content, applying a new stylistic constraint globally, and avoiding drift or partial compliance.

## 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`