Gap. Cost is fully observable after the fact (usage.py, live CostBadge) but nothing caps it — an orchestrator fanning out to specialists, each running Optuna and optional GPU sandboxes, can silently rack up spend.
Value. Essential trust infrastructure for an autonomous agent that spends real money.
Suggestion. Add a per-project/per-session budget (USD and/or token ceiling) that the runner checks against accumulated UsageEvent rows and pauses the agent with a request-clarification-style prompt when exceeded.
Gap. Cost is fully observable after the fact (
usage.py, liveCostBadge) but nothing caps it — an orchestrator fanning out to specialists, each running Optuna and optional GPU sandboxes, can silently rack up spend.Value. Essential trust infrastructure for an autonomous agent that spends real money.
Suggestion. Add a per-project/per-session budget (USD and/or token ceiling) that the runner checks against accumulated
UsageEventrows and pauses the agent with arequest-clarification-style prompt when exceeded.