Parse a CI test run and group failures by error signature. Deduplicates repeated errors and categorizes them as assertion, timeout, network, or crash. Use this as the first step before triage.

Given a list of test failures and a flakiness history, score each failure by how likely it is to be a known flaky test vs a real regression. Returns probability scores per test.

Match a list of changed files against failing tests to determine which failures are directly caused by the code changes in this commit. Returns a correlation mapping.

The final step: combines failures, flakiness history, and code change correlation to produce a GO / NO_GO / INVESTIGATE verdict with confidence score and a breakdown of blockers vs safe-to-ignore failures.

Analyzes a history of test failures with timestamps to detect chronometric patterns: failures that cluster at the same UTC hour (hourly jobs), same day of month (billing runs), same weekday (scheduled jobs), or around DST transitions. When a pattern is found, the failure is a time artifact — not a code regression. The agent should schedule a re-run at a different time rather than investigating the source code.

Scans a repository for versioned database migration files (Flyway V*.sql, Prisma migration.sql, Liquibase XML/YAML) and classifies each operation as additive (rollback safe) or destructive (forward-fix only). Returns rollback_eligible, a list of blocking_migrations with file and line, and deployment_strategy. Use before recommending deployment to determine whether a rollback is safe after go-live.