What problem does it solve?
E2E test suites in CI pipelines rely on manually maintained est_time values that drift out of sync with actual runtimes, causing poor partition balancing and wasted CI capacity. This Skill tunes those literal register_cuda_ci(est_time=...) values using real evidence from complete GitHub Actions run logs.
Core Features & Use Cases
- Evidence-backed runtime CSV: Parses complete job logs to produce an auditable CSV with per-file status, elapsed time, logged estimate, current estimate, and proposed estimate, citing exact log line numbers.
- Deterministic estimate computation: Aggregates passed samples per test file across all jobs and partitions, applies a 1.25 safety factor, and rounds upward to 10-second or 100-second buckets.
- Safe, estimate-only edits: Updates only literal
est_time values in single register_cuda_ci(...) calls, never touching test logic, and never staging, committing, or pushing changes.
- Use Case: After a nightly CI run, point the Skill at the run URL to recalibrate all
tests/e2e timing estimates, review the generated CSV of raises and lowers, and leave the verified edits uncommitted in the working tree.
Quick Start
Tune the E2E time estimates in tests/e2e using the complete logs from this GitHub Actions run URL and show me the resulting CSV.