Cascading triage
A cheap model classifies files as skip / skim / deep-review so the heavyweight model only runs where it pays off.
Rusty Bot
Cascading triage, OpenGrep pre-scan, and consensus voting — runs as a GitHub Action or webhook server.
Why Rusty Bot
Section titled “Why Rusty Bot”Tree-sitter context
Hunks expand to enclosing function/class boundaries — the model sees real semantic context instead of arbitrary line counts.
OpenGrep pre-scan
SAST findings on changed files inform triage before the LLM runs, so obvious security issues are never missed.
Consensus voting
Multiple review passes vote on each finding; lonely false positives are dropped automatically.
More to explore
Section titled “More to explore”Judge pass
A second model re-scores findings against a quality bar and drops the noise.
Convention files
Drop a REVIEW-BOT.md or .rusty-bot.md in your repo to teach the bot your
house rules.
Ticket integration
Pull Linear / Jira / GitHub issue context into the review so findings are judged against actual requirements.
Gating merges
Fail the check on critical findings to block merges until they’re addressed.
MCP tools
Connect MCP servers — internal docs, Sentry, custom tools — and let the review agent call them mid-review when it needs more context.