You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Reduce empty subagent polling with adaptive native waits #5
Outcome: publish a 60-second V2 minimum wait and five-minute default in all three arrangements, improve the existing optional orchestration skill and concise snippets, and document context flow, adaptive timing, limits and rollback. Owner: rafaelob. Acceptance: WHEN an arrangement is installed, THE catalog SHALL supply valid wait configuration and task-appropriate waiting guidance without changing budgets, goals, models, concurrency, permissions or hooks. Validate the public configuration contract and actual installed CLI configuration loading without paid inference; source inspection is not a token-savings benchmark. USER order in this conversation: implement the existing skill, concise AGENTS instructions and config changes, document, commit and push, and ask B to integrate the equivalent into the internal mechanism. Additional USER direction: choose waiting time compatible with task complexity and size. Next: implement and validate, publish through PR and CI, then hand off the internal adaptation to B. Rollback: restore only the changed wait fields and instruction sections.
Delivered in main commit 0fe823b via PR #6. Acceptance satisfied: native strict config/read loaded 60000/300000 for all three arrangements; invalid effective ordering rejected by the installed CLI. 39 tests and the Windows/Ubuntu Python 3.11/3.14 CI matrix passed. Evidence, limits and rollback are in docs/validation.md and docs/configuration.md. Budgets, goals, models, caps, permissions and hooks unchanged. No token/quota savings claim. Next: no remaining work for this issue.
Activity
feat: add adaptive native waits (#5) (#6)
rafaelob commentedon Sep 13, 2026
Delivered in main commit 0fe823b via PR #6. Acceptance satisfied: native strict config/read loaded 60000/300000 for all three arrangements; invalid effective ordering rejected by the installed CLI. 39 tests and the Windows/Ubuntu Python 3.11/3.14 CI matrix passed. Evidence, limits and rollback are in docs/validation.md and docs/configuration.md. Budgets, goals, models, caps, permissions and hooks unchanged. No token/quota savings claim. Next: no remaining work for this issue.