DrZero Analysis Mode

Read-only deep analysis of the codebase. Generates findings and recommendations without making any changes.

Installation Prerequisite

Install the DrZero Codex bundle before using this skill:

make codex-install-drzero SCOPE=repo

Analysis Dimensions

Architecture Review

• Dispatch agent: drzero-architecture
• Component boundaries and coupling
• Dependency graph analysis
• Design pattern identification
• Technical debt assessment

Security Audit

• Dispatch agent: drzero-security
• Vulnerability scanning (OWASP top 10)
• Secret detection in code and config
• Access control review
• Dependency audit (known CVEs)

Quality Assessment

• Code complexity metrics
• Test coverage gaps
• Documentation completeness
• Naming consistency

Performance Review

• Dispatch agent: drzero-performance
• N+1 query detection
• Resource utilization patterns
• Caching opportunities
• Bundle size analysis

Analysis Workflow

Analysis mode uses the DrZero proposer/solver pattern in read-only configuration:

1. Perspective Generation -- The drzero-proposer agent scans the codebase and generates analytical perspectives across five dimensions: architecture, security, performance, code quality, and technical debt. Each perspective becomes a discrete investigation task with a focused question and scope boundary.

2. Evidence Collection -- For each perspective, the drzero-solver agent (running in sandbox_mode: read-only) collects concrete evidence: file paths, line ranges, metric values, dependency chains, and pattern occurrences. No files are modified.

3. Multi-Perspective Synthesis -- Domain specialist agents are dispatched per dimension. The drzero-architecture agent evaluates component coupling and dependency graphs. The drzero-security agent checks OWASP categories and secret exposure. The drzero-performance agent profiles hot paths and resource patterns. Each specialist produces findings independently.

4. Scoring and Prioritization -- Findings use the same fail-closed scoring contract as the live Dr.Zero runtime. Proposer-side HRPO uses (0.5 * format_score + difficulty_score) / 1.5 (paper Section 3.2, rescaled to range [0.0, 1.0] from paper Eq. 4). Solver-side GRPO is only claimed when deterministic verification evidence exists; the canonical scorer derives a binary reward from observable outcome facts rather than model-authored utilities. Scores are computed via drzero-score.py, producing a ranked list of improvement opportunities with explicit provenance.

5. Report Assembly -- The orchestrator merges all specialist outputs into a single structured report. Duplicate or overlapping findings are consolidated. Cross-cutting concerns (e.g., a security issue that also affects performance) are linked.

Output

Generates a structured report with:

• Executive summary (3-5 key findings)
• Detailed findings per dimension
• Prioritized recommendations
• Estimated effort for top improvements

Use Cases

• Pre-refactoring assessment
• New team member onboarding
• Quarterly code health reviews
• Pre-release quality gates