santa-method

Verifies generated output through dual independent review agents with an iterative fix loop.

Updated Mar 25, 2026
One-click install
npx skills add https://github.com/Femad-6/my-skills --skill santa-method-femad-6
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: santa-method
Source: https://github.com/Femad-6/my-skills/tree/main/.github/skills/santa-method
Command: npx skills add https://github.com/Femad-6/my-skills --skill santa-method-femad-6

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? A single agent reviewing its own output shares the same biases and blind spots that produced the errors, so hallucinations, compliance violations, and inaccuracies slip into published or production deliverables. ## Core Features & Use Cases - Dual Independent Review: Two isolated reviewer agents evaluate output against an identical rubric, and both must pass before anything ships. - Convergence Loop: Flagged issues are fixed and re-reviewed by fresh agents each round, with a max-iteration cap and human escalation on exhaustion. - Batch Sampling Mode: Verifies a random 10-15% sample of large batches, classifies failure patterns, and applies targeted fixes across the whole batch. - Use Case: Before publishing AI-generated technical documentation, run both reviewers against a rubric covering factual accuracy, hallucination-free claims, and completeness, then iterate until both return PASS. ## Quick Start Ask the agent to verify this output using the Santa Method with two independent reviewers and a rubric covering accuracy, completeness, and compliance before shipping.

Frequently Asked Questions about santa-method

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify AI-generated content for hallucinations before publishing?

Run two independent review agents against the output using an identical rubric with objective pass/fail criteria for factual accuracy and fabricated references. Both reviewers must pass; otherwise fix the flagged issues and re-review with fresh agents until convergence.

How does dual-agent review work for quality assurance?

Two reviewer agents receive the same task spec, output, and rubric but share no context with each other. Each returns a structured JSON verdict with per-criterion results, and the output ships only when both return PASS.

When should I not use multi-agent adversarial review?

Skip it for internal drafts, exploratory research, and tasks with deterministic verification such as builds, tests, and linting. The 2-3x token cost is only justified for output that ships to users or carries compliance risk.

Can I use adversarial review on large batches of generated content?

Yes, use the batch sampling pattern: run full verification on a random 10-15% sample with a minimum of 5 items, classify failure patterns, apply targeted fixes to the whole batch, and re-sample until a clean sample passes.

Why do both reviewers keep finding new issues after fixes?

This is the infinite loop failure mode, mitigated by a maximum iteration cap of three rounds. After the cap is exhausted, the output escalates to a human reviewer instead of continuing the fix cycle.