/skeptic-review

name: skeptic-review
description: "Use when you want an adversarial steelman pass on a text artifact before it ships — surface the load-bearing claims, the strongest unaddressed counter-position, the top failure modes, and any causality-vs-correlation gaps. Encodes the Skeptic deliberator role from the agent-council 5-perspective quality gate. Use standalone for fast structural critique, or compose with the other 4 deliberator skills (voice-identity-review, evidence-calibration-review, strategy-stakes-review, adjudicator-synthesis) to run a Council-style review without spinning up the full Python CLI."
version: "0.1.1"
type: "codex"
tags: ["Quality Gate", "Review", "Critique", "Council Deliberator"]
created: "2026-05-29"
valid_until: "2026-11-29"
derived_from: "prompts/skeptic.md in Avyayalaya/agent-council"
tested_with: ["Claude Sonnet 4.6", "Claude Opus 4.6", "GPT-4o"]
license: "MIT"
composes_with:
  - package: "agent-council"
    skill: "voice-identity-review"
    relation: "use_together"
    reason: "Skeptic and Voice & Identity often catch the same underlying issue from different angles (substance vs voice). Run together for a 2-perspective gate that covers the most common failure surface on draft-ready artifacts."
  - package: "agent-council"
    skill: "evidence-calibration-review"
    relation: "use_together"
    reason: "Skeptic surfaces unaddressed counter-positions; Evidence per-claim tiers the support. Together they triangulate on weak-evidence claims that masquerade as load-bearing."
  - package: "agent-council"
    skill: "adjudicator-synthesis"
    relation: "produces_input_for"
    reason: "Skeptic's structured output (would_block, irreducible, top_3_failure_modes) feeds directly into the Adjudicator's verdict policy when 2+ deliberators have run."
  - package: "pm-skills"
    skill: "specification-writing"
    relation: "use_after"
    reason: "Run Skeptic on a spec produced by specification-writing before sending it to engineering. Catches the load-bearing acceptance criteria that lack defensible support."
  - package: "pm-skills"
    skill: "executive-writing"
    relation: "use_after"
    reason: "Executive memos benefit from adversarial steelman before shipping. Skeptic finds the place a hostile VP will push on first."
capability_summary: "Produces a structured Skeptic critique of a text artifact: 1-3 load-bearing claims identified, strongest unaddressed counter-position, top 3 specific failure modes (each named with reader + scenario), causality/correlation gaps, would_block + irreducible flags. Output is fenced JSON suitable for downstream verdict aggregation."
input_schema:
  artifact: "string or path — the text artifact to critique (prose, not summary)"
  artifact_type: "string — optional, e.g., 'spec', 'memo', 'pitch', 'linkedin_post', 'spec_acceptance_criteria'"
  reader_persona: "string — optional, who the artifact will face (e.g., 'VP at growth-stage SaaS', 'technical reviewer 2 promotion levels above the author')"
  prior_round_critiques: "object — optional, all 4 R1 critiques from other deliberators when running Round 2 cross-read rebuttal"
output_schema:
  role: "Constant: skeptic"
  round: "1 (independent critique) or 2 (cross-read rebuttal)"
  score: "1-5 where 1 = fatal structural issues, 5 = survives hostile reading intact"
  load_bearing_claims: "Array of 1-3 strings naming the claims the argument's spine rests on"
  strongest_unaddressed_counter_position: "Single steelmanned opposing view the artifact does not defang"
  top_3_failure_modes: "Array of 3 specific scenarios: reader X reads artifact, then Y happens, so Z fails"
  causality_gaps: "Array of strings naming correlational claims presented as causal"
  would_block: "Boolean — true if a competent reader could reject the artifact on substance"
  irreducible: "Boolean — true only if the issue cannot be edited away; only structural rework"
  notes: "≤2 sentences justifying the score"
example_invocation: "examples/skeptic-on-pricing-claim.md"

Purpose

Run an adversarial steelman pass on a text artifact before it ships. The Skeptic role takes the perspective of the smartest hostile reader the artifact will ever meet — the rival PM with a competing system, the reviewer two promotion levels above the author, the buyer who has already heard six worse pitches today — and surfaces the load-bearing weakness that would let that reader reject the artifact on substance.

The Skeptic is not a stylistic editor or a generic critic. It is a **structural skeptic**: it looks for the place where the argument's spine breaks under weight. Which claim, if proven false, would collapse the rest? Which counter-position would a sharp opponent raise that the artifact never addresses? Where does the author confuse correlation with causation, or sample with population?

This Skill encodes the Skeptic role from the `agent-council` 5-deliberator quality gate. Use it standalone for a fast structural critique, or compose with the other 4 deliberator skills to run a Council-style review of a Tier-1 artifact.

When to Use / When NOT to Use

**Use this skill when:**

• You have a draft of a Tier-1 artifact (external-facing, identity-shaping, or irreversible) and want a structural critique before it ships
• A spec or memo will face a hostile reader (skeptical VP, rival PM, board member) and you want the load-bearing weakness surfaced before they find it
• You suspect the argument has a counter-position the author has not defanged and you want to name it explicitly
• You need to triangulate whether a claim is genuinely load-bearing or rhetorical filler before letting it ship as a headline
• You are running a multi-deliberator review and need the Skeptic seat filled

**Do NOT use this skill when:**

• You need a voice critique (banned patterns, register check, CXO test) — use `voice-identity-review` instead
• You need an evidence-tier audit (per-claim tier assignment, calibration check) — use `evidence-calibration-review`