Waza Skills Development Platform - Product Requirements Document

June 28, 2026 · View on GitHub

Status: Active
Version: 1.0
Last Updated: 2026-02-06
Owner: @spboyer
Source: Squad Proposal


Executive Summary

Waza (技 - Japanese for "skill/technique") is a unified CLI platform for creating, testing, and evaluating AI agent skills. It consolidates existing skill development tools into a single binary that provides the complete developer experience for contributing to microsoft/skills.

The Problem

The microsoft/skills repository hosts 132+ skills for AI coding agents, but the contribution process lacks automated tooling for:

  • Compliance validation — No standardized scoring before PR submission
  • Trigger testing — Manual verification of skill activation patterns
  • Cross-model evaluation — No framework for testing skills across GPT-4o, Claude, etc.
  • Token budget enforcement — Guidelines exist but aren't automatically checked

The Solution

A single waza CLI built in Go that automates the skill development workflow:

PhaseCapability
ScaffoldGenerate compliant skill structure matching microsoft/skills conventions
DevelopIterate with real-time compliance scoring (Sensei engine)
TestRun agentic test loops with real LLM execution via Copilot SDK
EvaluateCross-model comparison with task completion, trigger accuracy, behavior quality metrics

User Personas

Primary: Skill Author

  • Role: Developer contributing skills to microsoft/skills
  • Goals: Create high-quality skills that pass CI, work across models
  • Pain Points: Manual testing, unclear compliance requirements, no cross-model validation

Secondary: Skill Reviewer

  • Role: Maintainer reviewing skill PRs
  • Goals: Quickly assess skill quality, ensure compliance
  • Pain Points: Inconsistent quality, manual verification

Tertiary: Platform Engineer

  • Role: CI/CD pipeline maintainer
  • Goals: Automate skill validation in pipelines
  • Pain Points: Lack of CLI tools for automation

Feature Requirements

Epic 1: Go CLI Foundation (P0)

Port existing Python waza functionality to Go for single-binary distribution.

IDUser StoryAcceptance Criteria
E1-01As a developer, I can run evaluations with waza runParses eval.yaml, executes tasks, outputs results
E1-02As a developer, I can initialize new eval suites with waza initCreates compliant directory structure
E1-03As a developer, I can create new skills with waza newScaffolds skill structure, supports --output-dir flag
E1-04As a developer, I can compare results across models with waza compareLoads multiple result files, generates comparison report
E1-05As a developer, I can use all 8 grader typescode, model, regex, file, keyword, json, script, composite
E1-06As a developer, I can execute against Copilot SDKFull integration with streaming responses
E1-07As a developer, I can use verbose mode for debuggingReal-time conversation display
E1-08As a developer, I can save transcripts for analysisJSON log output with full conversation
E1-09As a developer, public artifacts are schema-versionedeval.yaml and results.json include schemaVersion; readers warn on same-major drift and reject cross-major versions with a migration hint

Epic 2: Sensei Engine (P0)

Compliance scoring and iterative improvement loop.

IDUser StoryAcceptance Criteria
E2-01As a developer, I can run waza dev to start the improvement loopIterative scoring with feedback
E2-02As a developer, I can see my compliance score (Low/Medium/Medium-High/High)Clear scoring rubric applied
E2-03As a developer, I get specific improvement suggestionsActionable feedback per issue
E2-04As a developer, I can set a target scoreLoop until target reached
E2-05As a developer, I can run trigger accuracy testsshouldTrigger/shouldNotTrigger prompts
E2-06As a developer, I can skip integration tests with --skip-integrationUnit + trigger tests only
E2-07As a developer, I can use fast mode with --fastSkip tests for rapid iteration

Epic 3: Evaluation Framework (P0)

Cross-model testing and comprehensive metrics.

IDUser StoryAcceptance Criteria
E3-01As a developer, I can run evals against multiple modelsModel parameter support
E3-02As a developer, I can see task completion metricsPass rate, composite score
E3-03As a developer, I can see trigger accuracy metricsActivation pattern validation
E3-04As a developer, I can see behavior quality metricsResponse quality scoring
E3-05As a developer, I can run trials for statistical confidenceMultiple runs per task
E3-06As a developer, I can get LLM-powered improvement suggestions--suggestions flag
E3-07As a developer, I can run tasks in parallel--parallel flag
E3-08As a developer, I can filter to specific tasks--task flag
E3-09As a developer, I can reuse shared evals and graders from a registryRegistry design covers search/add/get UX, versioned refs, lockfile reproducibility, and safe plugin extensibility (design)
E3-10As a developer, I can verify eval coverage against SKILL.md requirementswaza spec verify maps description, trigger, anti-trigger, and parameter requirements to task coverage with CI-gateable output

Epic 4: Token Management (P1)

Budget tracking and optimization tools.

IDUser StoryAcceptance Criteria
E4-01As a developer, I can count tokens with waza tokens countToken count for all markdown files
E4-02As a developer, I can check limits with waza tokens checkValidate against budget
E4-03As a developer, I can use strict mode with --strictExit 1 if limits exceeded
E4-04As a developer, I can get optimization suggestions with waza tokens suggestLLM-powered reduction tips
E4-05As a developer, I can compare with previous commitswaza tokens compare HEAD~1

Epic 5: Waza Skill (P1)

Conversational interface for guided skill development.

IDUser StoryAcceptance Criteria
E5-01As a developer, I can use waza as a skill in CopilotSKILL.md published to microsoft/skills
E5-02As a developer, I get guided requirements gatheringInteractive prompts for skill creation
E5-03As a developer, I can check readiness conversationally"Is my skill ready?" triggers validation
E5-04As a developer, I get interpreted resultsPlain language explanation of scores
E5-05As a developer, the skill invokes CLI commandsSkill wraps waza CLI

Epic 6: CI/CD Integration (P1)

GitHub Actions and microsoft/skills compatibility.

IDUser StoryAcceptance Criteria
E6-01As a developer, I can run waza in GitHub ActionsAction workflow template
E6-02As a developer, I can fail PRs on low complianceExit codes for CI
E6-03As a developer, I can post results to PR commentsGitHub reporter output
E6-04As a developer, waza works with microsoft/skills CICompatible with existing test harness
E6-05As a developer, I can cache evaluation resultsIncremental testing support

Epic 7: AZD Extension (P2)

Package waza as an Azure Developer CLI extension.

IDUser StoryAcceptance Criteria
E7-01As a developer, I can install waza with azd extension install wazaPublished to azd extension registry
E7-02As a developer, I can run azd waza <command>All commands available via azd
E7-03As a developer, I get IntelliSense for waza commandsMetadata support
E7-04As a developer, waza integrates with azure.yamlConfiguration schema support

Technical Architecture

System Overview

┌─────────────────────────────────────────────────────────────────────────┐
│                           DEVELOPER WORKFLOW                            │
│                                                                         │
│   ┌─────────────────────────────────────────────────────────────────┐   │
│   │                          WAZA CLI (Go)                          │   │
│   │                                                                 │   │
│   │   init → generate → dev → run → compare                        │   │
│   └─────────────────────────────────────────────────────────────────┘   │
│                                    │                                    │
│              ┌─────────────────────┼─────────────────────┐              │
│              │                     │                     │              │
│              ▼                     ▼                     ▼              │
│   ┌──────────────────┐  ┌──────────────────┐  ┌──────────────────┐      │
│   │   Sensei Engine  │  │  Eval Framework  │  │   Waza Skill     │      │
│   │   (Compliance)   │  │ (Testing/Metrics)│  │   (Guidance)     │      │
│   └──────────────────┘  └──────────────────┘  └──────────────────┘      │
│                                                                         │
└─────────────────────────────────────────────────────────────────────────┘

Component Breakdown

ComponentLanguagePurpose
waza CLIGoMain binary, all commands
Sensei EngineGoCompliance scoring, improvement loop
Eval FrameworkGoTask execution, grading, metrics
Copilot ExecutorGoCopilot SDK integration
Waza SkillMarkdownSKILL.md for conversational interface

Directory Structure

/
├── cmd/waza/           # CLI entrypoint
├── internal/
│   ├── config/         # Configuration loading
│   ├── execution/      # Executors (mock, copilot)
│   ├── models/         # Data models (spec, task, outcome)
│   ├── orchestration/  # Runner, task coordination
│   ├── scoring/        # Graders, validators
│   └── sensei/         # Compliance engine (new)
├── go.mod
└── Makefile

Compliance Scoring System

Skills are scored on frontmatter compliance. Target: Medium-High or better for publishing.

ScoreRequirementsDescription
LowDescription < 150 chars OR no triggersBasic, agent can't route reliably
MediumDescription >= 150 chars AND has trigger keywordsFunctional but may have false positives
Medium-HighHas "USE FOR:" AND "DO NOT USE FOR:"Clear boundaries, reliable routing
HighMedium-High + INVOKES + FOR SINGLE OPERATIONSFull routing clarity, MCP integration

Success Metrics

MetricTargetMeasurement
Skill Compliance Rate>80% Medium-HighAutomated scoring via waza dev
Trigger Accuracy>90%Evaluation framework pass rate
Time to First Skill<30 minutesDeveloper surveys
Cross-Model Consistency>85% pass rate across 3+ modelsComparison reports
Token Efficiency<500 lines for SKILL.mdwaza tokens check

Dependencies

DependencyStatusRisk Level
Copilot SDKAvailableMedium - API stability
Go RuntimeStableLow
AZD Extension FrameworkAvailable (alpha)Medium - API changes
microsoft/skills repoExists (132+ skills)Low - Established

Phased Roadmap

Primary Phase (Core Features)

EpicDescriptionPriority
E1: Go CLI FoundationPort Python features to Go CLIP0
E2: Sensei EngineCompliance scoring & dev loopP0
E3: Evaluation FrameworkCross-model testing & metricsP0
E4: Token ManagementBudget tracking & optimizationP1
E5: Waza SkillConversational skill for guidanceP1

Secondary Phase (Integration & Extensions)

EpicDescriptionPriority
E6: CI/CD IntegrationGitHub Actions & microsoft/skillsP1
E7: AZD ExtensionPackage as azd extensionP2

Appendix: References

SourceDescription
microsoft/skillsTarget repository, contribution conventions
waza repoCurrent implementation
Squad ProposalOriginal proposal document
AZD Extension FrameworkExtension packaging guide