Learning
how often the agent agrees with you, and where it doesn't
Agent accuracy (30d)
—
no settled observations yet
Failures (30d)
0
false_close + missed_issue
Agreements (30d)
0
agreement_close + approve
Pending (30d)
20
not yet resolvable
Accuracy over time (last 90 days)
Recent misses
false_close + missed_issue · these drive prompt improvements
no misses yet — either the agent is perfect or there's not enough data
Pending prompt improvements
open PRs from the prompt-tuner agent · each one proposes a agent/prompt.md edit driven by the misses above
no open prompt-tuner PRs · the agent hasn't found enough failure cases to justify a prompt change yet