ACP Training Corpus SpecKit

A comprehensive Claude Code automation package for developing, validating, and managing the AI Context Protocol (ACP) training corpus.

Overview

This SpecKit provides end-to-end automation for creating a training corpus that embeds ACP knowledge into foundation models. The package follows the 4-phase implementation roadmap defined in the requirements and provides commands, scripts, and templates for each phase.

Quick Start

# Initialize the corpus project
/scaffold.init

# View current phase status
/validate.status

# Generate seed examples for a domain
/develop.seeds --domain annotation-syntax

# Run validation pipeline
/validate.pipeline

Package Structure

.claude/
├── commands/
│   ├── scaffold/           # Project initialization & setup
│   │   ├── init.md
│   │   ├── configure.md
│   │   └── baseline.md
│   ├── develop/            # Corpus development
│   │   ├── seeds.md
│   │   ├── expand.md
│   │   ├── multiturn.md
│   │   └── errors.md
│   ├── test/               # Testing & quality
│   │   ├── syntax.md
│   │   ├── semantics.md
│   │   └── behavioral.md
│   ├── validate/           # Validation & verification
│   │   ├── pipeline.md
│   │   ├── coverage.md
│   │   ├── status.md
│   │   └── requirements.md
│   ├── research/           # Research & analysis
│   │   ├── gaps.md
│   │   ├── competitors.md
│   │   └── resources.md
│   ├── implement/          # Implementation execution
│   │   ├── phase.md
│   │   ├── batch.md
│   │   └── review.md
│   └── remediate/          # Gap remediation
│       ├── identify.md
│       ├── plan.md
│       └── execute.md
├── memory/
│   ├── project-context.md
│   ├── phase-status.md
│   ├── coverage-matrix.md
│   ├── assumptions-log.md
│   ├── decisions-log.md
│   └── validation-state.md
scripts/
├── validate_syntax.py
├── validate_semantics.py
├── validate_schema.py
├── analyze_coverage.py
├── generate_seeds.py
├── check_requirements.py
├── estimate_resources.py
└── aggregate_metrics.py
templates/
├── examples/
│   ├── instruction-following.json
│   ├── multi-turn.json
│   ├── code-analysis.json
│   └── error-correction.json
├── documents/
│   ├── phase-report.md
│   ├── coverage-report.md
│   ├── validation-report.md
│   └── gap-analysis.md
└── reports/
    ├── daily-progress.md
    ├── phase-completion.md
    └── final-delivery.md
references/
├── requirements-traceability.md
├── corpus-categories.md
├── validation-rules.md
├── lock-level-semantics.md
├── cli-command-reference.md
├── mcp-tool-reference.md
└── ebnf-grammar.md
docs/
├── phase-1-foundation.md
├── phase-2-expansion.md
├── phase-3-completion.md
├── phase-4-validation.md
└── workflow-guide.md

Implementation Phases

Phase 1: Foundation (Weeks 1-2)

• Goal: 500 high-quality seed examples
• Focus: Lock levels, basic syntax, core CLI, primary MCP tools
• Commands: /scaffold.init, /develop.seeds, /validate.coverage

Phase 2: Expansion (Weeks 3-4)

• Goal: 2,500 examples + 500 multi-turn
• Focus: Variable system, multi-turn workflows, error correction
• Commands: /develop.expand, /develop.multiturn, /develop.errors

Phase 3: Completion (Weeks 5-6)

• Goal: 5,000 examples, >90% coverage
• Focus: Gap remediation, edge cases, integration scenarios
• Commands: /remediate.identify, /remediate.execute, /validate.pipeline

Phase 4: Validation (Week 7)

• Goal: Human review, behavioral testing, documentation
• Focus: Quality assurance, cross-validation, final delivery
• Commands: /implement.review, /validate.requirements, /implement.phase --complete

Requirements Traceability

All commands trace back to requirements in acp-training-corpus-requirements.md:

• Section 2: Corpus Objectives → /validate.requirements
• Section 5: Corpus Structure → /develop.* commands
• Section 9: Quality Gates → /validate.pipeline
• Section 11: Gap Analysis → /remediate.* commands
• Section 12: Implementation Roadmap → Phase commands

Memory Files

The SpecKit maintains persistent state across sessions:

Memory File	Purpose	Updated By
project-context.md	Project identity, stakeholders	/scaffold.init
phase-status.md	Current phase, progress metrics	All commands
coverage-matrix.md	Domain/category coverage tracking	/validate.coverage
assumptions-log.md	Tracked assumptions with validation	Manual + commands
decisions-log.md	Design decisions with rationale	Manual + commands
validation-state.md	Last validation results, issues	/validate.pipeline

Command Dependencies

scaffold.init
    ├── scaffold.configure
    │   └── scaffold.baseline
    │       ├── develop.seeds
    │       │   ├── develop.expand
    │       │   │   └── develop.multiturn
    │       │   └── develop.errors
    │       └── validate.coverage
    │           └── remediate.identify
    │               └── remediate.execute
    └── validate.pipeline (can run anytime)

Quality Gates

Every example must pass:

1. Syntactic Validation - EBNF grammar compliance
2. Semantic Validation - Lock level behavior, inheritance rules
3. Behavioral Validation - Workflow correctness, multi-turn consistency
4. Coverage Validation - Domain coverage, language distribution

Getting Started

1. Initialize: Run /scaffold.init to set up project structure
2. Configure: Run /scaffold.configure to set generation parameters
3. Baseline: Run /scaffold.baseline to establish initial metrics
4. Develop: Use /develop.* commands to generate examples
5. Validate: Use /validate.* commands to verify quality
6. Remediate: Use /remediate.* commands to fix gaps

Support

• Requirements: See acp-training-corpus-requirements.md
• Workflow Guide: See docs/workflow-guide.md
• Phase Details: See docs/phase-*.md