Prilizer: stop rereading every pull request

Описание и применение

Prilizer – локальный инструмент, который отслеживает до трёх pull‑request от одного автора, используя модель Qwen3 для выявления запросов и сохранения их до явного подтверждения. Он уменьшает необходимость повторного чтения PR, но сталкивается с задержками инференса.

Задача

Reviewers often reread pull requests after a new commit, wasting time when the request has already been addressed or the thread has moved on.

Результат

After refining the pipeline with source‑role and fresh‑confirmation constraints, the tool achieved 12/12 correct outcomes on synthetic cases, eliminating false "Waiting" results and invalid action quotes. However, model inference times ranged from 8.4 to 47.1 seconds, exceeding the 30‑second target.

Рабочий процесс

• 1. Pull request data (body, comments, reviews, inline comments) is fetched locally.
• 2. Qwen3 1.7B processes the data to propose structured requests.
• 3. Application validates quotes, provenance, actor roles, and freshness.
• 4. Validated requests are persisted independently of the next model summary.
• 5. The digest shows outstanding requests; users can mark as reviewed, dismiss locally, or confirm source.
• 6. If the model fails or data is partial, the card displays "Check manually" and retains older evidence.
• 7. Refreshes are manual; unchanged analyses are cached to avoid re‑analysis.

Доказательства

Prilizer is a local catch‑up desk for up to three public pull requests by one contributor.

Источник

The piece I care about most is what happens on the next refresh. An outstanding request stays visible even if the next summary omits it.

Источник

The first frozen evaluation failed: 8/12 correct outcomes, including one false Waiting.

Источник

The revised pipeline scored 12/12, with zero false Waiting results and zero invalid action quotes.

Источник

Fresh‑set model calls took 8.4–47.1 seconds. A real Docker CLI PR took 47.3 seconds, then reused its analysis on an unchanged refresh.

Источник