/insight-adversarial

Structured adversarial review that assembles a panel of distinct critic personas, runs each through a focused pass, and synthesizes a verdict. Personas are selected based on the nature of the target — a YON workflow gets different critics than a business plan or a UX design.

Structured execution spec: protocol.yon. Read it for the canonical rules and step sequence; this file is explanation. The two must stay in sync — if you edit one, update the other and refresh the @STAMP date.

Caller Options. Before executing, run the Caller Options protocol (caller-options/SKILL.md): triage this invocation for material optionality across the venues and modes declared in front-matter; if one path clearly dominates, proceed silently; otherwise surface the options to the caller. A resolved-invocation marker means COP already ran — execute the fixed combination directly, do not re-enter COP.

How Passes Work

State tracking: at the top of each pass, list which personas were used in prior passes. New passes must not reuse substantive critic personas; Current Reviewer's Own View recurs in every pass.

Step 1 — Target Classification

Before selecting personas, classify the target:

Load personas/roster.md to see all available personas and their best-used-for tags. Use the classification to select the most relevant 4–5 personas for this pass.

Step 2 — Persona Selection Logic

1. Pick the primary persona — whose domain most directly overlaps the target's failure mode
2. Pick 2–3 supporting personas — different angles that complement, not duplicate, primary
3. Always include Current Reviewer's Own View as the final persona in every pass
4. Never reuse a substantive critic persona across passes; Current Reviewer's Own View is exempt — track used personas explicitly

Step 3 — Run Each Persona Pass

For each selected persona:

1. State the persona name and their lens in one sentence
2. Report specific, evidenced findings. If none are supported, say no material finding supported in the inspected scope, naming that scope and remaining uncertainty. Never invent an issue to fill the format.
3. Merge convergent findings across personas, retaining their evidence; do not invent differences to make each lens unique.

Format:

## 🔴 [PERSONA NAME]
*Lens: [one-sentence description]*

[Critique — specific, no softening]

Current Reviewer's Own View always goes last. First person, clear position, no hedging.

Step 4 — Synthesis

• Critical Issues — up to 3 supported issues, ranked by actual severity; label lower severities honestly and say when no critical issue is supported. A sound target may have no material finding.
• Blind Spots — what the aggregate reveals that no single persona fully caught
• Remediation — minimum surgical fix per Critical Issue. If unfixable, say so.
• One-Line Verdict — honest, specific, no hedging

Step 5 — Pass Wrap Logic

• Pass 1 or 2: Ask user if they want another pass. If yes → load new personas, run delta critique focused on unresolved issues.
• Pass 3: Check if all Critical Issues resolved or explicitly accepted. If yes → final consolidated verdict + declare wrap. If no → flag remaining open issues, recommend escalation (e.g., /plan-create).

Rules

• MUST classify target type before selecting personas
• MUST select minimum 4 personas per pass (including Current Reviewer's Own View)
• MUST include Current Reviewer's Own View in every pass, always last
• MUST NOT reuse a substantive critic persona across passes; Current Reviewer's Own View is exempt
• MUST NOT offer solutions inside a persona pass — remediation belongs in Step 4 only
• MUST NOT soften findings to be diplomatic
• SHOULD add 3–4 genuinely new personas on Pass 2
• SHOULD tailor persona selection to target type
• MAY suppress a persona if user explicitly requests it

Files

• personas/roster.md — Full persona library. Load during Step 1 to inform selection.

Human output. This skill's handler-facing output obeys the human-output contract (human-output/SKILL.md).

Next skills. On completion, run the Next Skills protocol (next-skills/SKILL.md): surface the next-skills recommendations from front-matter for the caller to pick. Offer only — never auto-invoke.

Self-improvement. On completion, run the Self-Improvement Protocol (self-improve/SKILL.md): apply its signal taxonomy and threshold to this run; surface any surviving proposal for handler approval. Conservative — silent otherwise. Never auto-apply.