Skip to content
prashanth30-nPublic

About

This Project was made during the IARE Hactoberfest event

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Lore — turn a project's code-review history into an Agent Skill.

The problem. Every mature open-source project runs on unwritten rules: "use the logger, not print," "every user-facing change needs a changelog entry," "don't touch the public API without a deprecation." These rules live scattered across thousands of closed pull-request review comments. Maintainers type the same feedback over and over, new contributors learn the rules by getting their PRs rejected, and AI coding agents — which now write a growing share of contributions — can't follow rules they can't see.

What Lore does. You paste a GitHub repo link. Lore mines the project's closed PRs, review comments, issues, and commit history, then uses an open-weight model (via Ollama) to extract the rules maintainers actually enforce. It packages them as an Agent Skill (compliant with the Agent Skill Open Standard) that any compatible coding agent can load — and every rule cites the real review comments it came from, so maintainers can verify exactly where each rule originated. Alongside the skill, it generates a readable codebase overview: structure, key modules, entry points, and how to contribute — the "neat understanding of the codebase" new contributors need.

How it works. backend runs the pipeline: the miner pulls repo data through the GitHub API (file tree, READMEs, up to 50 closed PRs, review comments, commits, issues); the analyzer sends review history to the model in batches and extracts structured rules with citations, guarded by an invented-citation filter that drops any URL not present in the mined data; the skill generator emits a validated SKILL.md. A React frontend shows progress and presents three tabs — Overview, Rules (each with clickable citations to the real PR discussions), and Skill (preview + download).

Why it matters. Agent skills today are written by hand, so almost no project has one. Lore generates them automatically from the project's own history — turning tribal knowledge into something both humans and agents can actually use. For the Hacktoberfest challenge, the open-weight model is load-bearing: no model, no rules, no skill.

About

This Project was made during the IARE Hactoberfest event

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages