GitHub Action / Rust / BYOK

AI code review that actually reads your repo.

HoverStare reviews pull requests with repository context, read-only investigation tools, multi-pass voting, and precise inline comments that keep tracking findings across commits. In develop mode it also plans in the issue, opens the pull request, and keeps working from comments in that thread.

Add to a repo View source

Signal first

Built for reviews that need the rest of the codebase.

Repo-aware, not diff-only

The reviewing model can inspect files, grep call sites, and compare against the base branch before it reports a defect.

Voting plus verification

Independent passes look for correctness, concurrency, and security issues, then a verifier checks lonely findings before publication.

Precise inline comments

Reported line numbers are validated against the real diff and snapped to a valid anchor so comments land where reviewers expect.

Incremental by design

Hidden fingerprints follow findings across pushes, letting HoverStare review the delta and mark fixed issues without repeats.

Develop mode

The issue and the pull request are the IDE.

Discuss in the issue, hand it over with one command, and keep steering from the pull request. Every round commits to the branch, so the diff and the conversation stay in one place.

Plan, then build

@hoverstare in an issue starts a discussion round that investigates the repository. @hoverstare go turns the plan into a branch and a pull request.

Queue, not chaos

Instructions left on the pull request line up as tasks. One round runs at a time, the next stays queued, and an empty queue means no extra round.

Your name, your signature

Commits are authored by the person who asked, marked Co-authored-by: hoverstare[bot], and signed with your GPG key when you configure one.

Merges on command

@hoverstare merge checks the review state and the queue, then squash-merges and deletes the branch.

Fail-open Analysis trouble should not block CI.

BYOK Use Anthropic or an OpenAI-compatible endpoint.

Stateless Review state lives in GitHub comments and metadata.

Two minutes

Add HoverStare to a repository.

Create `.github/workflows/hoverstare.yml`, add an LLM key as a repository secret, and let pull requests trigger reviews.

name: HoverStare
on:
  pull_request:
    types: [opened, reopened, synchronize]

permissions:
  contents: read
  pull-requests: write
  statuses: write

jobs:
  hoverstare:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - uses: liuchong/hoverstare@v0.1.1
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          OPENAI_API_KEY: ${{ secrets.HOVERSTARE_LLM_KEY }}
          OPENAI_BASE_URL: ${{ vars.HOVERSTARE_LLM_BASE_URL }}
          HOVERSTARE_MODEL: ${{ vars.HOVERSTARE_MODEL }}

Commands

Talk to it in the PR.

@hoverstare review Force a fresh review when the conversation needs another pass.

@hoverstare explain Ask for a plain-language explanation inside a review thread.

@hoverstare go In an issue: implement the agreed plan on a branch and open a pull request.

@hoverstare <task> On a pull request: develop that task on the branch and report back.

@hoverstare queue Show what is running, what is still queued, and what failed.

@hoverstare merge Merge when checks are green and the queue is empty.

@hoverstare help Print the supported command list for the repository.

References

Read the details.

README Full setup, configuration, FAQ, and local development notes. Develop mode Issue-to-PR workflow, commands, commit identity and GPG signing. Pipeline How review passes, voting, verification, and reporting fit together. Serve mode Self-host HoverStare as a GitHub App webhook service. Specs The milestone plan and implementation contracts behind the project.