Argus PR Reviewer

Version updated for https://github.com/sudzxd/argus to version v0.2.0.

  • This action is used across all versions by ? repositories.

Action Type

This is a Docker action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

Argus is a GitHub Action designed to automate AI-powered code reviews by analyzing your entire codebase and providing precise, context-aware inline comments on pull requests. It solves the challenge of ensuring consistent code quality by learning your repository’s patterns and conventions, leveraging semantic mapping, hybrid retrieval methods, and multi-provider AI models. The action operates entirely within GitHub, requiring no additional infrastructure.

Release notes

Added

  • Incremental pattern analysis in index mode — Index mode can now optionally run LLM-based pattern analysis on changed files after updating the codebase map. Controlled by the INPUT_ANALYZE_PATTERNS env var (default "false"). When enabled, index pulls existing memory, runs incremental analysis scoped to changed files, and pushes updated patterns alongside the codebase map.
  • analyzed_at field on CodebaseMemory — Tracks the last commit SHA where patterns were analyzed, independent of indexed_at. Prevents history divergence between index and bootstrap modes.
  • Sharded artifact storage — Codebase map is split into per-directory shards with a DAG manifest (manifest.json) on the argus-data branch. Selective loading ensures reviews and index updates only fetch needed shards.
  • SelectiveGitBranchSync — Manifest-first pull with tree caching, selective blob download, and incremental push via base_tree merge.
  • LLM usage tracking — LLMUsage value object tracks token counts and request counts per strategy; reported in review logs.

Fixed

  • Outline preservation during incremental analysis — update_profile was replacing the full stored outline (all files) with a scoped outline (only changed files), causing memory.json to lose most outline entries. Now the scoped outline text is only used for LLM analysis while the full/existing outline is preserved in storage.
  • Bootstrap diff base — Bootstrap’s incremental path now uses analyzed_at (falling back to indexed_at) as the diff base for pattern analysis, so it correctly covers all changes since the last analysis rather than since the last index.

Changed

  • CI workflow — argus-index.yml incremental index step now sets INPUT_ANALYZE_PATTERNS: "true" to enable pattern analysis on push.