# Improve Judgment Hygiene

Judgment-heavy answers often blur facts, assumptions, recommendations, and missing evidence. This skill gives Claude, Codex, and Claude Code internal checks for cleaner decisions.

## Install

```bash
npx skillstore add syntagmanull/judgment-hygiene
```

## Metadata

- Status: approved
- Slug: syntagmanull-judgment-hygiene
- Skillstore revision: r1
- Version status: missing
- Tree hash: 8a536ad19270ad87bb1cd0533252abc3894d8e53214ec3bc7f17d62ddca0e714
- Author: SyntagmaNull
- GitHub username: SyntagmaNull
- License: MIT
- Repository: https://github.com/SyntagmaNull/judgment-hygiene-stack/tree/main/judgment\_hygiene
- Ref: 36e07d5e13068e5be64447e8f20b427cf2cbd21a
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: confirmation\_required
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands
- Quality score: 38
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/syntagmanull-judgment-hygiene
- Manifest: https://skillstore.pages.dev/api/skills/syntagmanull-judgment-hygiene/manifest

## Capabilities

- Separates observations, inferences, evaluations, actions, unknowns, and tradeoffs in judgment-bearing answers.
- Guides certainty levels so weak evidence is not presented as strong proof.
- Checks that evaluations are grounded in specific observations or inferences.
- Adds meaningful tradeoffs to nontrivial recommendations.
- Selects honest abstention modes when evidence is incomplete.
- Resists irrelevant framing changes that could distort a judgment.

## Use Cases

- Evaluate product decisions: Compare options while separating evidence, assumptions, risks, and tradeoffs before making product recommendations.
- Clarify ambiguous incidents: Turn unclear incidents into grounded answers that distinguish observed facts from likely causes and unresolved gaps.
- Improve AI answer quality: Add judgment checks to Claude, Codex, or Claude Code workflows that need careful recommendations.

## Prompt Templates

### Basic judgment check

```
Use judgment hygiene on this question. Separate what is given, what is inferred, and what remains uncertain: [question or context].
```

### Recommendation repair

```
Review this draft recommendation for hidden assumptions and missing tradeoffs: [draft]. Return a cleaner version.
```

### Ambiguous situation analysis

```
Analyze this ambiguous situation. Identify direct evidence, plausible interpretations, evaluation criteria, and what could change the conclusion: [situation].
```

### Pipeline final answer

```
Apply judgment hygiene after structure_judgment and verification_hygiene. Use the routing context and evidence payload to produce a concise final answer: [context].
```

## Limitations

- It does not verify facts or retrieve external sources by itself.
- It is internal guidance, not a required visible output format.
- It cannot guarantee correct moral, legal, medical, or financial conclusions.
- It depends on the base model applying the checks consistently.

## Best Practices

- Use the skill when an answer includes a recommendation, evaluation, diagnosis, or interpretation.
- Keep structure mostly internal unless exposing uncertainty or tradeoffs improves clarity.
- Pair it with source verification for factual claims that need current evidence.

## Anti Patterns

- Using labels as decoration without changing the answer's evidence structure.
- Treating every uncertain point the same instead of showing real confidence differences.
- Giving nontrivial advice without naming meaningful costs, risks, or prerequisites.

## Security Audit

- Audited at: 2026-07-07T07:40:46.337\+00:00
- Summary: All static findings are false positives caused by Markdown inline-code backticks, ordinary example fences, and ordinary reasoning prose. I found no executable command, system reconnaissance, network reconnaissance, or data exfiltration behavior in SKILL.md. One semantic issue remains: the skill includes unverified approval language that could influence trust decisions.

## Stats

- Views: 121
- Downloads: 11
- Favorites: 0
- Popularity score: 0