English | 简体中文

Practical tools for AI-assisted software delivery.

I build tools around the everyday challenges of working with AI agents: keeping projects moving across conversations, making responsibilities clear, keeping documentation useful, and getting commands to execute correctly.

Keep complex AI projects moving through to delivery.

Five collaboration levels, L0–L4, help organize work from a single task to a project spanning multiple phases. Clear roles cover coordination, planning, execution, and independent review.

Model and reasoning defaults are configurable. They are informed by my private evaluations built from real repositories and tailored to each role's responsibilities.

Keep documentation in step with the project.

Audit, organize, maintain, and recover project documentation. Clarify where decisions and current state belong, preserve useful existing structures, and help documentation evolve with the software it describes.

Help AI agents get Windows commands right.

Handle the practical boundaries between PowerShell, native executables, text encoding, paths, and WSL. Diagnose failures and verify what a command actually completed.

• Session Coordinator — Connect related sessions, preserve coordination records, and support recovery after interruptions.
• Work Charter for DSH — Bring Work Charter's roles, decisions, evidence, and recovery model into DeepSeek Harness, using Session Coordinator for cross-session coordination.

Both plugins currently have public GitHub pre-releases. Installation instructions and compatibility details live in their repositories.

• Start with recurring problems in real projects.
• Keep tools focused and responsibilities clear.
• Make results verifiable and interrupted work recoverable.
• Evaluate defaults, then refine them through actual use.

Each project is maintained independently. Visit its repository for documentation, source, releases, and current status.