Project context memory for AI coding agents

The problem

AI coding agents start every session with a blank slate. You waste tokens and time re-explaining your directory structure, specific tech stack versions, and coding conventions every time you open a new chat.

What it does

• Scans your repository to identify languages, runtimes, package managers, and core dependencies automatically.
• Generates a PROJECT_CONTEXT.md file that maps your directory structure and detects linting, testing, and naming conventions.
• Maintains a persistent decision log to track architectural changes and migration history over time.
• Updates the context file incrementally at the end of sessions to reflect new dependencies or folder path changes.
• Preserves manually written notes in protected sections while automating the boilerplate repo mapping.

Frameworks & tools

Compatible with any repository using standard manifests like package.json, pyproject.toml, Cargo.toml, go.mod, or Gemfile. Works across Claude Code, Cursor, Aider, Windsurf, and Cline.

Why this beats prompting it yourself

Manually keeping a project map updated is tedious and usually falls out of sync. This skill enforces a strict schema that ensures any agent, regardless of the IDE, immediately understands your specific environment without hallucinating your project structure.

Use cases

• Onboarding an AI agent to a complex legacy codebase for the first time.
• Handing off context between different AI tools, such as moving from Cursor to Aider.
• Documenting major architectural pivots in the decision log for future reference.
• Enforcing consistent linting and style conventions across multiple AI-generated PRs.