lore

Stop losing critical project context to the "memory hole" of long AI chat sessions. Lore provides a framework-agnostic, long-term memory layer for your codebase that persists across agents, IDEs, and sessions. It captures the high-level architecture, design decisions, and coding conventions that live in a senior developer's head but rarely make it into the documentation.

What it does

• Persistent memory stores project architecture, decisions, and conventions in plain Markdown files within a .lore/ directory.
• Bi-directional syncing records new decisions or architectural changes after features, refactors, or bug fixes.
• Platform mirroring projects project memory into agent-specific files like CLAUDE.md, .cursorrules, or AGENTS.md.
• Intelligent auditing compares stored project memory against the actual source code to identify drift and stale information.
• Context compression generates a SUMMARY.md digest to help agents quickly ingest large knowledge bases without hitting token limits.

How it works

1. Initialize the memory bank by scanning the codebase and workspace configuration to establish the initial .lore/ structure.
2. Record changes using sync to capture the "why" behind non-trivial code modifications and update relevant entries.
3. Recall context via query to answer questions about project conventions or historical design choices.
4. Maintain accuracy through periodic audit and compress cycles to ensure memory matches reality and stays concise.

Frameworks & tools

This skill is language and framework agnostic. It supports monorepo detection for pnpm, Yarn, npm, Lerna, Nx, Rush, Cargo, Go, and Bazel. It integrates with Claude Code, Cursor, Aider, and Cline via mirror files.

Why this beats prompting it yourself

Manual prompts about project context are transient and inconsistent across sessions. This skill enforces a deterministic ID system and a structured ARCH/DEC/CONV schema that ensures your project's history remains searchable, auditable, and transferable between different AI agents.

Use cases

• Record why a specific library was chosen over a popular alternative to prevent future refactoring loops.
• Enforce custom linting or architectural rules that generic AI models don't know by default.
• Onboard a new AI agent to a complex monorepo by providing a structured summary of every scope.
• Audit a legacy codebase to find where the actual implementation has drifted from documented conventions.

Known limitations

Does not replace changelogs or version control. Requires manual confirmation for file mutations. Mirror updates are not triggered on every sync by default to avoid git noise.