Lore: A Structured Way to Feed Your Team's Past Decisions to Coding Agents

Every AI coding agent eventually asks the same question your team already answered six months ago. Lore is an open-source CLI tool that captures those decisions — architectural choices, naming conventions, dependency preferences — and feeds them to agents as structured context.

The core idea: instead of stuffing a giant markdown file into your repo’s .cursorrules or CLAUDE.md, Lore stores decisions as discrete, queryable records it calls “lore entries.” Each entry has a category, a rationale, and metadata about when and why the decision was made.

What it does concretely:

• Captures decisions via CLI — lore add walks you through recording a team decision with structured fields (category, description, rationale, alternatives considered)
• Serves context to agents — a local MCP server exposes lore entries so coding agents can query relevant decisions mid-task instead of guessing
• Organizes by category — architecture, conventions, dependencies, testing patterns, and custom categories
• Stores everything locally — JSON files in your repo, version-controlled alongside code

The project surfaced on Hacker News with 37 points and 48 comments, drawing a mix of interest and skepticism. The recurring debate: is this meaningfully different from a well-maintained ARCHITECTURE.md or a set of cursor rules files? The Lore team’s argument is that flat files don’t scale — agents pull the entire file into context whether or not it’s relevant, burning tokens and diluting signal. Structured entries let the agent query only what it needs for the current task.

That’s a real problem. Anyone who’s watched a coding agent hallucinate a dependency choice their team explicitly rejected knows the pain. But the value proposition depends entirely on whether teams will actually maintain lore entries over time — the same adoption problem that kills most internal documentation tools.

Early-stage and open source — the repo is public on GitHub, MIT-licensed, and actively taking feedback. Worth watching if your team keeps re-explaining the same decisions to AI agents.