Add judgment-hygiene-stack skill for lightweight framing, verification, and action checks

[SyntagmaNull]

Summary

This PR adds a new Codex skill: judgment-hygiene-stack.

The skill is designed as a lightweight judgment cleanup tool, not a full reasoning framework. It helps catch a small set of common failure modes in judgment-heavy prompts:

• treating user framing as fact
• stretching local evidence into global conclusions
• skipping needed verification
• over-verifying prompts that are emotional but not externally checkable
• recommending actions without surfacing cost, risk, or reversibility
• forcing evidence into the user's framing when the verified result is actually orthogonal

Why this skill

Many hard prompts do not fail because the model cannot generate an answer. They fail because the answer starts from the wrong layer, inherits a loaded premise, verifies the wrong claim, or gives advice without enough structural discipline.

This skill is meant to be small and composable:

• it does not replace judgment
• it does not force visible reasoning structure
• it does not wrap ordinary prompts by default

Instead, it offers a short set of internal checks that can be applied only when needed.

What is included

• SKILL.md with a short trigger-oriented entrypoint
• references for:
  • structural routing
  • verification hygiene
  • judgment hygiene
  • examples

Intended use

Use this skill for prompts involving:

• premise-smuggling
• local-to-global judgment jumps
• current or source-sensitive factual claims
• meaningful action recommendations
• orthogonal verification results
• emotionally loaded prompts where the model may overvalidate or become artificially cautious

Do not use it for:

• simple direct factual tasks
• formatting or rewriting
• straightforward retrieval
• ordinary coding tasks

Design note

This skill is intentionally scoped as a lightweight corrective tool. It is not intended to serve as a general-purpose reasoning method or a replacement for model judgment.
This skill is useful because it targets recurring, model-agnostic judgment failures without imposing a heavy framework. It is narrow enough to be practical, but broad enough to help across relationship, policy, legal, medical, and action-oriented prompts where structural drift is common.
The skill is especially aimed at preventing confident but structurally wrong answers, not at making answers longer or more cautious by default.