Skeptic

skeptic

A token-efficient sanity check, tripwire anomaly detection, and assumption validation skill for AI coding agents.

Why

AI agents frequently execute multi-step pipelines on unverified assumptions—such as spending days processing duplicate databases before realizing both sources were identical. `skeptic` introduces calibrated skepticism: **zero token overhead** on routine edits, **passive tripwires** for statistical anomalies, and **scalar probes** (<20 tokens) for high-blast-radius operations.

Skill Structure

skeptic/
├── SKILL.md                  # Core skill rules, tripwires, and gate patterns
├── references/
│   ├── scalar-probes.md      # 1-line bash/SQL recipes (<20 tokens)
│   └── tripwires.md          # Anomaly signatures directory
├── README.md
└── LICENSE

Installation

Using `npx skills` (Recommended)

Install directly using the universal Agent Skills CLI:

npx skills add IyedNasra/skeptic

Manual Installation

Clone into your agent's skills directory:

• Claude Code:

  git clone https://github.com/IyedNasra/skeptic.git ~/.claude/skills/skeptic

• Antigravity / Gemini CLI:

  git clone https://github.com/IyedNasra/skeptic.git ~/.agents/skills/skeptic

• Cursor / Custom Agent Runtimes: Clone into .cursor/skills/skeptic or include SKILL.md in your project instructions.

Core Principles

• The Asymmetry Rule: Low-risk routine changes proceed with zero verification overhead. High-blast-radius operations execute a 1-line scalar gate.
• Zero Context Pollution: Uses compact scalar probes (hashes, line diffs) to prevent thousands of tokens of raw data from flooding the context window.
• Tripwire Anomaly Sensing: Passive awareness for anomalies (duplicate sizes, instant execution, zero-variance outputs) that trigger verification only when something smells off.

License

[MIT](LICENSE)