# Doubt

> Doubt is an open-source Agent Skill, zero-dependency Node.js CLI, and GitHub
> Action for turning contested questions into source-grounded evidence maps.

Repository: https://github.com/alsoleg89/doubt
License: MIT
Current release: https://github.com/alsoleg89/doubt/releases/tag/v0.8.0

## Core contract

- Atomic position, claim, evidence, and unknown nodes.
- Typed supports, contradicts, qualifies, and missing reasoning edges.
- Every evidence node requires a dated source, bounded excerpt, and exact
  section, page, timestamp, or line locator.
- Decorative sources, unsourced evidence, dangling edges, unsupported
  positions, and invented confidence percentages fail validation.
- Valid maps receive a SHA-256 receipt over canonical JSON and recorded source snapshots.
- Explicit `doubt verify` retrieval can match excerpts and bind receipts to
  retrieved-byte hashes; partial failures produce no attested output.
- Validation and rendering are local-only; there is no telemetry or model API.

## Public artifacts

- Product and map gallery: https://alsoleg89.github.io/doubt/
- Local-only playground: https://alsoleg89.github.io/doubt/playground/
- Agent Skills portability map: https://alsoleg89.github.io/doubt/examples/agent-skills-portability.html
- Agent Skills vs MCP map: https://alsoleg89.github.io/doubt/examples/agent-skills-vs-mcp.html
- Reader benchmark: https://alsoleg89.github.io/doubt/benchmark/
- Five-client portability protocol: https://github.com/alsoleg89/doubt/blob/v0.8.0/benchmarks/skill-portability/README.md
- Codex fixed-skill result: https://github.com/alsoleg89/doubt/blob/v0.8.0/benchmarks/skill-portability/results/codex-0.146.0-alpha.3.1-offline.json
- GitHub Copilot result: https://github.com/alsoleg89/doubt/blob/v0.8.0/benchmarks/skill-portability/results/github-copilot-1.0.77.json

## Installation

Agent Skill:

    gh skill install alsoleg89/doubt doubt

CLI:

    npx doubt-ai validate decision.doubt.json
    npx doubt-ai verify decision.doubt.json --out decision.verified.doubt.json
    npx doubt-ai map decision.doubt.json --out decision.html

GitHub Action:

    uses: alsoleg89/doubt@v0.8.0

## Scope limits

The validator proves structural traceability and safe rendering. It does not
prove that a source is true, that an AI extracted it faithfully, or that a
verdict is correct. The Agent Skills portability benchmark has submitted runs
for two of five clients: Codex and GitHub Copilot. These results do not establish
cross-client behavioral equivalence.