CC Plugin Decision Log

Why it's ranked here

The design targets a costly agent failure: forgetting settled choices and repeating failed debugging attempts. Searchable project history and automatic compaction recovery form a coherent workflow. The experimental warning matters, but the integration, hook, edge-case, and end-to-end tests provide credible engineering evidence.

What's good

Decision records include considered options, the chosen answer, rationale, and optional tags. Problem records distinguish failed and successful attempts, preserve details, and capture final resolutions. Open problems retain full approach history during compaction, while resolved problems receive concise summaries. Data stays in readable local JSON.

Tradeoffs

The author calls it experimental and says real Claude Code usage remains sparse. Recording depends on the agent consistently logging choices and each attempted approach. Problem histories do not persist across sessions like decisions do. Storage uses whole-file synchronous JSON reads and writes without file locking, which creates scaling and concurrency concerns.

How to use it well

Use it for long Claude Code sessions involving architecture choices, debugging branches, or frequent context compression. Log a decision when alternatives matter later, and record every troubleshooting attempt before moving onward. It complements source control and project documentation, but does not replace either or provide general task management.

Observed

Primary language

TypeScript

Interface

MCP server using standard input and output transport

Tool surface

Seven MCP tools for decisions, problem tracking, search, listing, and context recovery

Install surface

Claude Code plugin directory loading or repository-based marketplace configuration

Runtime

Node.js launches the compiled server

Storage

Local JSON files grouped by project and session under the Claude user data directory

Testing

Repository includes integration, hook, edge-case, and end-to-end workflow test files