Workload Analysis

The Workload Analysis Layer inspects the developer’s workspace context, active files, project structure, language, and complexity, to build a Workload Profile that describes what kind of work the request entails.

What Is a Workload Profile?

A workload profile is a structured snapshot of the task at hand. It includes:

Field	Description	Source
`language`	Detected programming language(s)	File extension + heuristics
`stack`	Framework or runtime identified	`repo/intelligence/profiler.ts`
`complexityLevel`	Estimated complexity (1-5)	`complexityLevel()` in workload/analyzer.ts
`contextBucket`	Size of context window needed	Message length + attachment size
`outputBucket`	Expected response length	Keyword prediction in workload/signals.ts
`executionPriority`	urgency × complexity × cost sensitivity	`deriveExecutionPriority()` in workload/analyzer.ts
`confidence`	How certain the analysis is	`computeConfidence()` in workload/analyzer.ts

The Analysis Pipeline

Classified Intent
    |
    v
[1] Scan Active Files     --> Detect language, stack, file count
    |
    v
[2] Assess Complexity     --> complexityLevel(), classifyComplexity()
    |
    v
[3] Measure Context       --> contextBucket, outputBucket estimates
    |
    v
[4] Compute Priority      --> deriveExecutionPriority()
    |
    v
[5] Build Workload Snapshot -> buildWorkloadSnapshot()

Complexity Assessment

The complexity engine looks at:

Message length, longer prompts suggest more complex tasks
File attachments, .ts, .tsx, .py, .rs files carry different complexity weights
Multi-step indicators, keywords like “and then”, “after that”, “in addition” indicate multi-step reasoning
Domain keywords, “architecture”, “refactor”, “design pattern” score higher than “variable name”

Signal Functions

Key functions in workload/signals.ts:

Function	Purpose
`extractComplexityScore()`	Scans request text for complexity keywords
`extractLatencySensitivity()`	Detects urgency cues (“quick”, “now”, “ASAP”)
`extractCostSensitivity()`	Detects budget-conscious language (“cheap”, “cost”, “save”)
`estimateContextSize()`	Measures message + attachment tokens
`predictOutputLength()`	Predicts response length from keywords (“short”, “detailed”, “comprehensive”)
`predictToolUsage()`	Detects agentic language implying tool calls
`detectWorkflowStyle()`	Classifies as interactive, agentic, or batch
`computeCodingWeight()`	Weights coding-specific signals higher than general chat

File Reference

File	What It Does
`workload/analyzer.ts`	Main analysis orchestrator. Builds the complete workload profile
`workload/signals.ts`	Signal extraction library. Pure functions for text analysis
`workload/types.ts`	TypeScript types for workload profiles, signals, and buckets
`repo/intelligence/profiler.ts`	Repository-level profiling. Detects monorepo structure, identifies stack
`repo/intelligence/scanner.ts`	File system scanner. Walks workspace to detect languages and frameworks

Integration with Downstream Layers

The workload profile feeds into:

Strategy Engine, workload complexity determines which strategy is selected
Provider Scoring, context bucket size filters out models with small context windows
Economics, cost sensitivity from workload analysis caps the budget ceiling
Execution, execution priority determines timeout profiles and retry policies