# Synthesize Text with a Multi-LLM Council

Long documents can be hard to summarize without losing meaning or biasing the result. This skill compares multiple model perspectives and records the reasoning trail.

## Install

```bash
npx skillstore add alexmacapple/synthese-multi-llm
```

## Metadata

- - Slug: alexmacapple-synthese-multi-llm
- - Version: 1.0.0
- - Author: Alexmacapple
- - GitHub username: Alexmacapple
- - License: MIT
- - Repository: https://github.com/Alexmacapple/Synthese-Council/tree/main/skills/synthese-multi-llm
- - Ref: main
- - Supported tools: Claude, Codex, Claude Code
- - Risk level: medium
- - Risk factors: scripts, network, filesystem, env\_access, external\_commands
- - Quality score: 50
- - Quality tier: warning
- - Public page: https://skillstore.pages.dev/skills/alexmacapple-synthese-multi-llm
- - Manifest: https://skillstore.pages.dev/api/skills/alexmacapple-synthese-multi-llm/manifest

## Capabilities

- Runs Claude, Gemini, and Codex as expert reviewers for the same source text.
- Assigns extraction, fidelity review, and structure roles to different model perspectives.
- Supports standard, quick, critical, pedagogical, and direct synthesis modes.
- Computes a convergence score and highlights disagreement between model outputs.
- Saves audit trails and optional Markdown summaries for later review.
- Falls back to fewer available models when the full council is not installed.

## Use Cases

- Review a policy brief: Create a balanced summary and identify points where meaning could shift during compression.
- Compare research interpretations: Ask several model roles to extract claims, critique omissions, and converge on a final synthesis.
- Prepare executive summaries: Turn a long report into a targeted summary with audience, tone, and length controls.

## Prompt Templates

### Summarize a Document

```
Use synthese-multi-llm to summarize this document for a general audience. Keep the key claims and note major uncertainties.
```

### Create an Executive Brief

```
Use the council to summarize this report for executives. Make it concise, formal, and focused on decisions.
```

### Check Semantic Drift

```
Compare this source text with my draft summary. Identify omissions, additions, and shifts in meaning before revising it.
```

### Run a Traceable Council Review

```
Run a standard multi-model synthesis with audit trail enabled. Preserve dissenting points and explain the convergence score.
```

## Limitations

- Requires installed and authenticated model CLIs or a configured backend.
- May send source text to external model providers during normal use.
- Summary quality depends on the selected models and the source document quality.
- Audit trails may store source excerpts locally unless trail saving is disabled.

## Best Practices

- Review source sensitivity before sending documents to external model providers.
- Use the critical mode when accuracy matters more than speed.
- Disable audit trails when local persistence of source text is not acceptable.

## Anti Patterns

- Do not process secrets, credentials, or confidential records without an approved data policy.
- Do not treat the convergence score as proof that the summary is correct.
- Do not run configurable CLI backends from untrusted configuration files.

## Security Audit

- - Safe to publish: true
- - Audited at: 2026-06-28T08:17:42.469\+00:00
- - Summary: Static analysis found many command, credential, network, filesystem, and hash patterns. Review confirms these are mostly intended multi-LLM orchestration features, not confirmed malicious behavior. The skill should publish with a medium-risk warning because it can send source text to model providers and persist audit data locally.

## Stats

- - Views: 184
- - Downloads: 6
- - Favorites: 0
- - Popularity score: 0