RFC: optional proactive Saved Reset policies

[korkin25]

Hi @masterkain — before expanding a code patch, I would appreciate your thoughts on this optional Saved Reset behavior.

My main goal is to keep an agent's work moving when its current account approaches quota exhaustion. If a reset is available, I would like Pooler to redeem it early enough for the quota to recover, including during a long-running turn, rather than letting the work stop first. I understand that provider latency and in-flight behavior prevent an absolute continuity guarantee.

The proposal has two independently enabled policies, both off by default:

• Protect the account actually serving work near its limit, even if another pool account has spare capacity. An idle account would not trigger this policy.
• Rescue an available reset close to expiration without requiring traffic on that account, retaining reserve and consumption safeguards.

I would also like per-account overrides with optional deployment/Helm defaults. The attached document describes the intended behavior and safeguards; it contains no runtime changes.

Does this direction fit the project? Would you prefer extending the existing modes or separate switches? In particular, which quota/permission signals and active-turn integration would you recommend for timely redemption without wasting credits? If you prefer a smaller first contribution or a different configuration interface, I can adapt the implementation accordingly.

I have paused expansion of a local prototype while seeking your feedback. I need a working setup, but would much rather build this in a form you would accept upstream than maintain a competing implementation.

Validation: proposal-only diff reviewed locally; no live policies changed and no reset credits consumed.