phi-analyzer

# phi-analyzer

Auto-invoked compositional project analysis using φ = ∫(structure × semantics × memory).

## Description

This skill is invoked when agents need project context. Analyzes codebase using three integrated layers: structure (deterministic maps), semantics (curated annotations), and memory (cross-session learnings). Use for understanding project layout, architecture, and accumulated insights.

## Trigger Conditions

Invoke automatically when:
- User mentions "project structure", "codebase overview", or "architecture"
- Agent starts task requiring project awareness
- Fresh session in directory with `.phi/` folder
- Commands like "explain how X works" or "find where Y is implemented"
- Building features that need architectural context

## What It Provides

**Layer 1: Structure (Deterministic)**
- File locations and organization
- Module exports and imports
- Language breakdown
- AST-level structure from PROJECT-MAP.auto.scm

**Layer 2: Semantics (Curated)**
- Architecture patterns (e.g., JEA layers)
- Module purposes and relationships
- Known issues and technical debt
- Trust boundaries and security model
- From PROJECT-MAP.scm

**Layer 3: Memory (Learned)**
- Cross-session insights from vessel
- Relief-guided patterns that worked
- Architectural decisions and rationale
- Previous debugging learnings

## Capabilities

- ✓ Compositional filesystem queries via mcp__periphery__discover
- ✓ S-expression map parsing and analysis
- ✓ Vessel memory semantic search
- ✓ Progressive disclosure (index → details on-demand)
- ✓ Fantasy Land combinators for transformation pipelines

## Safety

**Low-risk** - Read-only analysis, no code modifications. Safe to auto-invoke.

## Output Format

```
φ Project Analysis
══════════════════

Structure (176 files, 85% TypeScript)
├─ jurisdictions/ - Smart contracts (J layer)
├─ runtime/ - BFT consensus (E layer)
└─ frontend/ - 3D visualization (A layer)

Architecture: JEA (Jurisdiction-Entity-Account)
  J: On-chain dispute settlem