Runtime Intelligence

Runtime Intelligence connects agent activity to the configuration and pull request involved.

Questions it answers

1. Which Harness Pack and provider were available?
2. Which capabilities were actually used?
3. What failed or required permission?
4. What happened to the resulting pull request?

Use the Runtime view to inspect configuration adoption, local drift, tool or skill usage, failures, verification activity, and observed review outcomes.

Read comparisons carefully

Fullbeam reports observed associations, sample sizes, and source coverage. It does not claim that a prompt, skill, model, agent, or engineer caused an outcome.

For example, Fullbeam may report that a pack version appeared in PRs with fewer review rounds. That is evidence worth investigating, not proof that the pack caused the difference.

Unavailable data is not the same as zero. Fullbeam labels missing or partial sources so that absent telemetry is not presented as no activity.

Improve the harness

Use repeated evidence to decide what deserves a controlled change:

• a capability is assigned but never used;
• a tool repeatedly fails;
• a verification step is missing from PRs that need rework;
• a pack version is associated with a changed review outcome.

Create and publish a new Harness Pack revision rather than modifying an already published version.