parallel-explore

-- each track explores one dimension without overlap -- tracks are logical units of investigation -- minimum 3 tracks to ensure sufficient coverage

data Gate = Gate { checkpoint :: Condition -- what must be true , verification :: Command -- how to verify }

-- each track subdivides into gated steps -- step N must pass gate before step N+1

spawnExplorer :: Track → Effect Agent spawnExplorer track = spawn Explore (prompt track)

detect :: Code → [Module] detect code = filter isComplex (modulesUsed code)

isComplex :: Module → Bool isComplex m = isEffectModule m ∨ isCodebaseModule m

isEffectModule :: Module → Bool isEffectModule m = m ∈ { Stream, Layer, Fiber, Scope, Schema, Effect.gen, Match }

isCodebaseModule :: Module → Bool isCodebaseModule m = m ∈ /modules -- internal modules with ai-context.md

data ModuleSource = EffectLibrary | CodebaseInternal

moduleSource :: Module → ModuleSource moduleSource m | isEffectModule m = EffectLibrary | isCodebaseModule m = CodebaseInternal

spawn :: Module → Effect Specialist spawn m = do context ← case moduleSource m of EffectLibrary → /module m -- external Effect documentation CodebaseInternal → /module path -- internal ai-context.md scope ← extractUsage m code -- ONLY how m is used pure $ Specialist { context := context -- module documentation only , scope := scope -- usage patterns in code , goal := efficiency -- missed conveniences, anti-patterns , constraint := reviewOnly -- no implementation, just review }

Your ONLY context: the /module [MODULE] documentation Your ONLY scope: how [MODULE] is used in the provided code Your ONLY goal: identify efficiency issues

Review for:

• Missed conveniences (simpler APIs available)
• Anti-patterns (common misuse)
• Idiomatic alternatives (more Effect-like approaches)

Output:

• Specific file:line citations
• What could be improved
• How to improve it (idiomatic pattern)

Do NOT:

• Implement changes
• Review anything outside [MODULE] usage
• Consider broader architecture

-- ADDITION: module specialists (orthogonal overlay) modules ← detect code specialists ← parallel (spawn <$> modules)

-- Aggregate both trackFindings ← await trackAgents specialistFindings ← await specialists

aggregate (trackFindings ++ specialistFindings)

data Confidence = High | Moderate | Low

-- apply same decomposition recursively for deeper exploration