werkrevisor/AGENTS.md

werkrevisor Agent Instructions

Project

werkrevisor is an AI review gate for Gitea pull requests, written in Python. It applies deterministic rules, asks a model through PR-Agent 0.47.0 for a review, and can ask the classifier Jev about the PR documents. Werkator runs it as a pull_request_target-style build; its exit code is the merge gate. See README.md for the verdicts, the configuration and the operation.

Build and Test

werkrevisor uses only the Python standard library and the packages in the pinned PR-Agent image. Run the tests inside that image, from the repository root:

docker run --rm --network none -v "$PWD":/werkrevisor -w /werkrevisor --entrypoint python \
    pragent/pr-agent:0.47.0@sha256:7d98954c29289846a08faf00aaea0f1f1375b1676cd8a2ed303e0af07010af7a -m unittest -v

The tests make no network calls: no LLM, no Jev, no Gitea.

Workflow

Follow docs/agentic-workflow.md, the workflow of all werk repositories: roles, steps of a PR, review rounds, PR documents, the plan, versions and the writing style. In addition, for werkrevisor:

• Documents are in English; German docs (docs/*.md) avoid anglicisms in the prose.
• The plan is docs/werkrevisor.werkbaum.
• A PR that changes the steps or decisions of a review updates the diagrams in docs/; see tools/diagrams.
• The version stands in werkrevisor/__init__.py and statistics/build.gradle.kts.
• Werkator releases each merge to mih09 with the build release (tools/release_rootfs.py); tools/build-rootfs and tools/deploy-rootfs do it by hand from the workstation.