Checkie

A Claude Code skill that reviews AI responses through multiple skeptic lenses. Get a critical second opinion before trusting AI output for important decisions.

Core Personas

Every review runs through 5 specialized skeptic personas:

Occam

Logic and Reasoning — cuts through fallacies, unsupported claims, circular reasoning, contradictions

Receipts

Evidence and Accuracy — demands proof for unverified facts, cherry-picking, outdated info, misleading stats

Cassandra

Completeness and Blind Spots — sees what others miss: missing perspectives, edge cases, unstated assumptions

The Mirror

Bias and Framing — reflects confirmation bias, sycophancy, framing effects, cultural assumptions

Murphy

Practicality and Real-World Fit — if it can go wrong, it will: feasibility, hidden costs, failure modes

Custom Skeptics

User-defined — create your own skeptic profiles with custom names, focus areas, and review prompts tailored to your domain

Modes

Mode	Command	Description
Full	`/check`	All personas, full output, actionable suggestions
Quick	`/check quick`	Condensed single-table summary
Critical	`/check critical`	Only surfaces 3-star-or-below concerns
Compare	`/check compare`	Side-by-side comparison with an alternative
Ask	`/check ask [question]`	5 personas answer a question independently
Export	`/check export`	Save review as markdown or HTML file
Auto	`/check auto on`	Auto confidence score after every response

Installation

# Download and install
curl -LO https://checkie.ai/skill/checkie-skill.tar.gz
mkdir -p ~/.claude/skills/checkie ~/.claude/commands
tar xzf checkie-skill.tar.gz -C ~/.claude/skills/checkie
cp ~/.claude/skills/checkie/commands/*.md ~/.claude/commands/

Slash Commands

Command	Description
`/check`	Skeptical review
`/doublecheck`	Skeptical review
`/op2`	Skeptical review
`/second-opinion`	Skeptical review

Features

Actionable Suggestions

Every review ends with 3-5 numbered improvements. Reply with a number to apply fixes automatically.

Confidence History

Scores tracked over time. Ask "show my checkie history" to see trends and weak spots.

Multiple Modes

Quick summaries, critical-only filtering, side-by-side comparisons, and independent persona answers.

Auto Mode

Enable with /check auto on to get a quick confidence score after every AI response.

Requirements

Any Claude client (Claude Code, Claude Desktop, Claude Chat, Cowork, Antigravity)
No external dependencies — runs entirely as a prompt-based skill