Add cost accounting and per-task cost aggregation.
Track cumulative cost across all agentd runs for a given task, so we can see how much each task costs to develop/review/integrate.
Implement:
1. After each agentd run completes, extract cost from agentd status <run-id> --json (check what fields are available).
2. Add a task comment with the run cost: "Run dev-t-575-... cost: $X.XX"
3. Optionally add a task cost <id> CLI subcommand that sums all cost comments for a task.
4. Add cumulative cost to the loop's log output.
5. Add --max-task-cost CENTS flag: if cumulative cost for a single task exceeds this, stop retrying and mark as needs-human-attention.
This is lower priority since the provider is buffet-priced, but useful for understanding system efficiency and detecting runaway tasks.
Files: Omni/Ide/dev-review-release.sh, optionally Omni/Task.hs
Implemented per-run cost accounting and cumulative task budget cap in commit e4204bcf.
Changes:
cost_cents= and cumulative metadata--max-task-cost guard for per-task cumulative spendneeds-help with explanatory commentManual review due recurring CostExhausted in workflow review runs. Verified with shellcheck/bash syntax and runtime probes: status/dashboard still works, cost comments are emitted after runs, and path now escalates to needs-help when limit exceeded.
Manual review due recurring CostExhausted in workflow review runs.
Verified:
bash -n Omni/Ide/dev-review-release.shlint Omni/Ide/dev-review-release.shstatus + status --json)--max-task-cost) escalates task to needs-help and avoids further automation attemptsIntegrated commit e4204bcf onto live as e38a181b (manual integration path due provider budget exhaustion in workflow agents).
Implemented per-run cost accounting and cumulative task budget cap in commit e4204bcf. Added run cost comments (, cumulative), loop logging, guard, and needs-help escalation when over budget; docs updated.