popup-studio-ai/bkit-claude-code

What it solves

AI-generated code often lacks precision, drifts from original requirements, or exceeds the context limits of LLM sessions. This project addresses the "coding gap" where AI produces plausible but incorrect code that doesn't match the user's intent or design specifications.

How it works

Built as a plugin for Claude Code, bkit uses a "Context Engineering" approach to manage complex development tasks. It employs a structured workflow including:

• Sprint Management: Splits large releases into context-budgeted sprints (e.g., ≤ 75K tokens) to prevent memory overflow and allow sessions to resume where they left off.
• PDCA Loop: Implements a Plan-Do-Check-Act cycle using specialist agents (PM, CTO, QA, etc.) to handle everything from requirements gathering to final reporting.
• Quality Gates: Uses a gap-detector to measure the match rate between design specs and generated code. If the match rate falls below 90%, an auto-repair mechanism attempts to fix the code.
• Autonomy Control: Provides a "Trust Level" dial (L0 to L4) that allows users to decide how much of the workflow runs unattended.

Who it’s for

• First-time vibe coders: Provides a safety net that measures AI output against specs and auto-repairs discrepancies.
• Solo developers: Acts as a "team-in-a-box" by spawning parallel specialist agents for frontend, backend, and security.
• Team leads: Enables sprint management by splitting releases into manageable, context-aware chunks.
• Non-English speakers: Supports 8-language auto-detection to trigger commands.

Highlights

• Self-repairing code: Automatically iterates to close the gap between design and implementation.
• Context-aware memory: Uses task management to ensure progress survives across different coding sessions.
• Multi-agent orchestration: Spawns specialized teams for product management, architecture, and QA.
• Automated documentation: Follows a "Docs = Code" philosophy where every feature generates a full audit trail of plans and reports.