Metrics

Hours from the first coding session to PR merge or close — shorter means faster delivery from idea to shipped code.

Merged PRs per contributor per week — higher means the team is shipping more frequently with agent assistance.

How many commits were pushed after the PR was opened — fewer means the agent got it right the first time.

How many back-and-forth turns it takes to finish a task — fewer means your prompts are clear and the agent stays on track.

Highest percentage of the model's context window used in any single message — high values mean the session is pushing against limits.

How much work the agent does between human interventions — higher means it operates independently with less hand-holding.

Fraction of tool calls that are slash commands or custom MCP tools — higher means the team is leveraging advanced agent capabilities.

Fraction of tool calls that delegate to subagents — higher means the agent is parallelizing work effectively.

Fraction of PRs with 50+ changed lines merged within 5 minutes of opening — high rates suggest insufficient code review.