PR-Agent changelog: what's new each month
Every stable PR-Agent release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 3 months covered; the current month updates daily.
October 2026 · so far
1 release: 0.47.0
October brought several security and reliability improvements, including stricter webhook validation, better error handling in GitHub Actions, and enhanced performance for Bitbucket Server webhooks.
Action needed
- Gitea webhooks require a configured secret; set gitea.webhooksecret to enable webhooks.
- GitHub Action now fails on tool errors by default; set githubactionconfig.failontoolerrors = false to restore previous behavior.
- Comment commands on GitLab, Bitbucket, and Bitbucket Server must start with a slash to be recognized.
- extraconfigurl can no longer be set in repository .pragent.toml files; configure it at the host level instead.
Highlights
- Gitea webhooks now require a configured secret for security.
- GitHub Action now fails visibly when tools encounter errors instead of silently continuing.
- Comment commands on GitLab, Bitbucket, and Bitbucket Server now require a leading slash.
- Bitbucket Server webhook container now runs under gunicorn with multiple workers for better isolation.
Improved
- Bitbucket Server webhook container now uses gunicorn with multiple worker processes instead of a single Python process.
- GitHub Action error handling is stricter, causing the workflow to fail on tool errors by default.
Fixed
- Gitea webhooks are now rejected with HTTP 403 when no secret is configured, preventing unsigned webhook acceptance.
- Comment-based commands on GitLab, Bitbucket, and Bitbucket Server now only trigger when prefixed with a slash.
September 2026
2 releases: 0.45.0 → 0.46.0
September brought improvements to review handling and cross-repository functionality, along with support for new AI models and better error recovery. The month focused on making PR-Agent more flexible for complex codebases and reducing review friction.
Highlights
- Review findings now persist across reruns, reducing duplicate feedback.
- New per-directory configuration support for monorepo projects.
- Automatic commands now acknowledge pull requests immediately when starting.
- Support added for Gemini 3.8 Flash and Grok 4.3 reasoning models.
- Repository context can now be loaded from sibling repositories in the same namespace.
New
- Opt-in large-diff chunking with merge rule configuration for handling large changes.
- Trigger PR-Agent commands after review submission on GitHub.
- Content-based auto-resolve for fixed inline threads on GitLab.
- Configurable description-based issue matching for tickets.
- Typed prompt output models for structured responses.
- Public Git provider registration API.
- CodeCommit support for reviewing all pull request targets.
- Fallback model support to recover failed analysis chunks.
Improved
- Suggestion capping per file after chunk aggregation to reduce noise.
- Optional timeout handling with same-model retry skip.
- Better monorepo support through sibling repository context loading.
Fixed
- Failed analysis chunks now recover using fallback models instead of failing completely.
August 2026
4 releases: 0.41.1 → 0.44.0
August brought expanded model support, improved platform integrations, and several breaking changes to deployment and development workflows. New features focused on compliance checking, inline feedback, and AI reasoning capabilities.
Action needed
- GitHub App webhooks require GITHUB.WEBHOOKSECRET to be configured; unsigned webhooks are rejected with HTTP 403.
- Dependency management migrated to uv; requirements.txt, requirements-dev.txt, and setup.py removed.
- Linting now uses Ruff exclusively; flake8 and standalone isort pre-commit hook removed.
- LiteLLM upgraded to v1.98.0; verify compatibility with your configuration.
Highlights
- GitHub App webhooks now require a configured secret for security; deployments without GITHUB.WEBHOOKSECRET will be rejected.
- Dependency management migrated from requirements.txt to pyproject.toml with uv; setup.py is removed.
- Added support for multiple new models including Claude adaptive thinking, Gemini 3.7 Flash, and Grok 4.5/4.6.
- Azure and GitLab platforms now support inline review findings and incremental review/improve commands.
- New opt-in agent run details section shows model, tokens, time, and AI call counts.
New
- Support for /improve command on local git provider.
- Optional provider-side request attribution via OpenAI user field.
- Asana ticket detection added to ticket compliance check.
- Agent run details section showing model, tokens, time, and AI call metrics.
- GLM-5.2, Kimi K3, Qwen3.8, and MiMo v2.5 model support.
- Feedback on draft PRs to trigger auto commands.
- OpenRouter router model support.
- Claude adaptive thinking support via LiteLLM.
Improved
- GitLab can now publish /review output as resolvable threads.
- Review coverage now surfaces files excluded from review.
- Changes diagram direction made configurable and adaptive.
- GitLab /review and /improve commands now support incremental mode (-i flag).
- Azure review findings now published inline.
- LiteLLM pass-through cache control injection points for Anthropic prompt caching.
- Max reasoning effort now configurable.
- Dependency pins relaxed to patched ranges for easier downstream security updates.
Fixed
- LiteLLM now surfaces real provider errors instead of RetryError.
- Local git provider falls back to old path when diff item path is None.
- Per-request settings cloning fixed in Bitbucket Server and Azure DevOps webhooks.
- Azure span complete suggestion lines in review feedback.
- Stale progress comments removed from suggestions.
- Suggestion line spanning corrected for Azure.
Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. PR-Agent: pricing, features and alternatives · All changelogs