Define standing-context re-delivery policy for provider compaction
- Dominant language
- Rust
- Stars
- 32.7k
- Forks
- 4.3k
- Avg merge
- 1d 13h
- Merged PRs (30d)
- 253
Description
## Problem
Legacy ACP sessions now receive standing context once per live Buzz session. If the provider compacts that context away while the session remains live, Buzz has no signal or threshold that causes the persona/base/system context to be sent again.
## Desired outcome
Choose and implement a policy for legacy providers, such as:
- re-deliver standing context after a turn/token threshold;
- configure a non-zero session rotation default; or
- use a provider capability/signal when available.
The policy must avoid restoring the old behavior of repeating all standing context on every turn.
Follow-up from #5342 session-context delivery work; this is a product/session-lifecycle decision rather than a defect in the bounded delivery-state change.
Contributor guide
Research direction
Start by reviewing the session-context delivery work in #5342 and the legacy provider session-lifecycle entry points. Compare threshold-based re-delivery, session rotation, and provider capability options; done means a documented and implemented policy that restores compacted standing context without repeating it on every turn.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- rust
- Domain
- backend
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Needs clarification
- Newbie friendliness
- 30/100