Repository navigation
0.20.0 — Dry-run reviews & inline prompt overrides
This release adds a dry-run mode for iterating on prompts and models against real PRs without posting, and lets prompt env vars accept either a file path or raw content directly — closing the loop between "edit a prompt" and "see what a review looks like." A config reshape separates bot identity from agent runtime config.
What's new
- Dry-run mode — run the full review pipeline end-to-end without posting comments to the PR and without writing to the conversation store. The review still produces a
ReviewResultfor inspection. Useful for testing prompt/model changes against existing PRs non-destructively.