fix(skills): restore Aristotle Integration in adversarial skill

PR #26's rewrite of the adversarial skill collapsed the explicit "Aristotle
Integration" section into a one-line mention, dropping the canonical phase
names. This regressed TestAdversarialReferences (test_has_aristotle_
integration_section, test_references_assumption_autopsy) — masked in CI by a
`|| true` on the pytest step, so it merged silently. It also violates the
non-negotiable "preserve Aristotle + 5 phases" project principle.

Restore a dedicated "## Aristotle Integration" section listing the five
canonical phases (Assumption Autopsy, Irreducible Truths, Reconstruction,
Assumption vs Truth Map, Aristotelian Move), turned toward opposite-analysis.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: alfredolopez80

.claude/skills/adversarial/SKILL.md

@@ -23,7 +23,15 @@ Use when the user says `adversarial`, `analisis opuesto`, `opposing analysis`, `
 5. Separate proven issues from plausible risks.
 6. Recommend one concrete next action: keep, adjust, defer, validate, or revert.
 
-For non-trivial work, apply a compact Aristotle pass: assumption autopsy, irreducible truths, reconstruction from evidence, assumption-vs-truth map, and the smallest risk-reducing next move.
+## Aristotle Integration
+
+For non-trivial work, run a compact Aristotle First Principles pass — the same five phases the orchestrator uses — turned toward the opposing position:
+
+1. **Assumption Autopsy** — list every assumption the claim or plan depends on.
+2. **Irreducible Truths** — strip the assumptions; keep only what survives in the current repo/runtime.
+3. **Reconstruction from Zero** — rebuild the strongest counter-position from those truths alone.
+4. **Assumption vs Truth Map** — separate what is proven from what is merely assumed.
+5. **The Aristotelian Move** — the single highest-leverage next action: keep, adjust, defer, validate, or revert.
 
 ## External Advisor Mode