retro

The retro skill serves as an advanced diagnostic tool for senior engineers and CTOs to audit team velocity and individual contribution patterns. By parsing git history through Claude Code, it translates raw commit data into actionable insights, helping teams move from mere output tracking to understanding depth of focus, shipping consistency, and architectural health. It is particularly effective for teams using Conventional Commits, as it automatically categorizes work into feature, refactor, and fix types, flagging high-bug ratios that might suggest systemic technical debt or insufficient code review cycles.

• Performs automated work session detection using a 45-minute idle threshold to categorize time spent into deep, medium, or micro-sessions.

• Generates Pacific-time-aware hourly commit distribution histograms to identify peak productivity windows and late-night burnout patterns.

• Calculates a custom Focus Score based on directory-level commit concentration, helping identify if engineering energy is directed or scattered.

• Performs automatic hotspot analysis to detect churn-heavy files, highlighting potential bottlenecks and versioning discipline.

• Aggregates PR size distributions to flag oversized changes that risk stability.

• Tracks shipping streaks and historical trends when windowing reaches 14+ days.

• Trigger this skill using the /retro command with arguments like 7d, 24h, 30d, or compare for window-based analysis.

• Ensure the repository is up to date with git fetch origin main, which the tool handles internally before parallelizing log processing.

• Uses TZ=America/Los_Angeles for all timestamp conversions; ensure your environment reflects this for accurate data presentation.

• Ideal for senior ICs needing objective data to support architectural pivots or workload balancing.

• Flags XL PRs (1500+ LOC) to promote better scoping and smaller, safer code deployments.