Feature request from a voice/chat operating session.
Problem: agent_loop_runs already stores, per run, the objective, how many iterations it actually took (iteration/max_iterations), and how it ended (status/stop_reason: done, failed, max_iterations, cancelled, etc). But nothing rolls this history up into a usable view. Right now the only way to see a run's outcome is to read individual run records one at a time via check_work/coding_loop_status — there's no aggregate.
Requested behavior: a per-instance stats view (new read tool and/or a console dashboard panel) built from the existing agent_loop_runs table, showing at minimum:
- Success rate: proportion of runs that ended 'done' vs 'failed' vs 'max_iterations' vs 'cancelled', per instance.
- Step-count distribution: how many iterations runs actually took before finishing, so it's visible whether an instance/objective type tends to finish early, or routinely burns to its cap.
- Ideally viewable per instance, so each coding agent's own track record is visible on its own (e.g. 'this agent finishes done in a median of 6 steps' vs 'this agent hits max_iterations 40% of the time').
This is read-only reporting over existing data — no new run-time behavior, no changes to how the loop itself runs. Just make the historical outcome data collected in agent_loop_runs queryable/visualizable in aggregate instead of only one row at a time.
Use case that surfaced this: wanting a rough steady-state read on each coding agent — is it usually completing cleanly, and in roughly how many steps, or is it routinely running out of budget — without manually reading through run after run.
Feature request from a voice/chat operating session.
Problem: agent_loop_runs already stores, per run, the objective, how many iterations it actually took (iteration/max_iterations), and how it ended (status/stop_reason: done, failed, max_iterations, cancelled, etc). But nothing rolls this history up into a usable view. Right now the only way to see a run's outcome is to read individual run records one at a time via check_work/coding_loop_status — there's no aggregate.
Requested behavior: a per-instance stats view (new read tool and/or a console dashboard panel) built from the existing agent_loop_runs table, showing at minimum:
This is read-only reporting over existing data — no new run-time behavior, no changes to how the loop itself runs. Just make the historical outcome data collected in agent_loop_runs queryable/visualizable in aggregate instead of only one row at a time.
Use case that surfaced this: wanting a rough steady-state read on each coding agent — is it usually completing cleanly, and in roughly how many steps, or is it routinely running out of budget — without manually reading through run after run.