NavanithanS

ask-conceptual-integrity-sentinel

Audit repos for architectural drift, dead code, and abstraction bloat.

NavanithanS 1 1 Updated 3mo ago

Resources

4
GitHub

Install

npx skillscat add navanithans/agent-skill-kit/ask-conceptual-integrity-sentinel

Install via the SkillsCat registry.

SKILL.md
❌ NO hallucinating purpose of ambiguous code → flag as "High Entropy" ❌ NO suggesting more abstraction → suggest removal/simplification ❌ NO guessing untraceable calls → mark "Untraceable - Refactor Required" ✅ MUST surface assumptions explicitly ✅ MUST generate SENTINEL_REPORT.md ✅ MUST provide "Burn List" (top 3 files too clever) </critical_constraints> - Assumption Drift: verify imports, don't assume from package.json - Confusion Management: 500-line utils.js → flag, don't hallucinate - Abstraction Bloat: 10-line logic in 3 class layers → "Premature Optimization" </failure_modes> 1. Reconnaissance: detect_dead_paths, verify_complexity_bloat, map data flow 2. Interrogation: apply Simplicity Filter to every suggestion 3. Report: generate SENTINEL_REPORT.md ❌ Bad: "Refactor into Strategy Pattern" (adds weight) ✅ Good: "Delete Factory, use direct function call" (removes weight) </simplicity_filter> ## SENTINEL_REPORT.md - **Slop Index**: % of dead/redundant code - **Critical Assumptions**: what I assumed (risks if wrong) - **Burn List**: top 3 "too clever" files - **Architecture Gaps**: missing tests, circular deps, no error boundaries </output_format> - Low complexity/LOC ratio + large file → boilerplate/slop - High complexity/LOC ratio + small file → code golf/unreadable - Prop drilling >5 layers → architecture smell