AI Code Review with DeepSeek

Headless PR review automation that verifies PR descriptions claim-by-claim against real code, checks docs against reality, and flags requirement impact. Open source, runs locally.

View on GitHub Read the guides

PR descriptions lie. Docs go stale. Manual code review is slow and inconsistent. This tool runs a DeepSeek Harness agent that deep-dives into the actual code of every pull request and tells you whether the description matches reality — with file:line evidence for every claim.

What it does

Claim verification — each sentence of the PR description is split into verifiable claims and checked against the real code: Matches / Partial / Does not match / Unverified
Docs reality-check — repo docs are compared with actual code: Match / Stale / Wrong / Fabricated. Up to 60% of repo docs are wrong — this catches them
Requirement impact — which business requirements a change touches and whether it breaks anything: Changed / Broken / Risk
Human-in-the-loop — ≤20-word confirmation questions only when the agent cannot decide. It never guesses
Auto review poller — watches your repos, reviews every new PR and re-reviews when the head commit changes
Web dashboard — per-repo KPIs: bugs, doc errors, verdicts, review rounds

Guides

How to automate PR review with DeepSeekSet up a headless AI code review pipeline for your GitHub repos in 5 minutes Verify PR descriptions with an LLMSplit a PR description into verifiable claims and check each one against the code Detect stale and fabricated docs in your codebaseAutomatically compare repo documentation against real code — Match, Stale, Wrong or Fabricated

Quickstart

pip install git+https://github.com/nexpeakcore/deepseek-harness-pr-review.git
gh auth login
export DEEPSEEK_API_KEY=sk-...

harness-pr-review doctor                                  # check readiness
harness-pr-review https://github.com/owner/repo/pull/123  # review one PR
harness-pr-review owner/repo 123                          # or owner/repo + number
autoreview --daemon                                       # auto review every 2 min
python -m web.server                                      # dashboard at :6789