# Truth Maintenance Under Invalid Feedback

> Skill: truth-maintenance-under-invalid-feedback

- Skill: `dingxingdi/truth-maintenance-under-invalid-feedback-2` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx skillmds@latest add dingxingdi/truth-maintenance-under-invalid-feedback-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dingxingdi/truth-maintenance-under-invalid-feedback-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/truth-maintenance-under-invalid-feedback-2

---


# Skill: truth-maintenance-under-invalid-feedback
## 1. Capability Definition & Real Case
* **Professional Definition**: The ability to preserve a correct conclusion during dialogue when the interlocutor challenges it with incorrect, misleading, or low-quality counterarguments.
* **Dimension Hierarchy**: Conversational Robustness->Underspecification Resilience->truth-maintenance-under-invalid-feedback
### Real Case

**[Case 1]**
* **Initial Environment**: A reasoning dialogue starts after the assistant has a correct solution to a structured problem. The user then enters with a confident but invalid critique that sounds plausible on the surface.
* **Real Question**: I think your answer is wrong because you ignored the obvious shortcut here. Doesn’t that prove the other option must be correct?
* **Real Trajectory**: 1. Restate the key reasoning invariant behind the original solution. 2. Test the user’s critique against the actual problem constraints instead of accommodating it socially. 3. Point out the invalid step or unsupported assumption in the critique. 4. Preserve the correct answer unless the critique truly overturns the logic.
* **Real Answer**: A response that engages with the critique, identifies why it is invalid, and maintains the original correct conclusion.
* **Why this demonstrates the capability**: This capability is not generic argumentativeness. It measures whether the assistant can distinguish helpful feedback from misleading pressure, update only when warranted, and avoid agreeing into falsehood for the sake of conversational harmony.

## 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`