Improvement queue
Auto-drafted fixes from conversations that did not go well — review each one, then accept and apply it or dismiss it.
The improvement queue (/improvements) turns failed conversations into concrete, reviewable fixes. Syncanix drafts each proposal; you decide whether to apply it.
What lands in the queue
Proposals come in three kinds, each drawn from a conversation that did not resolve well:
- Tool description — a clearer or corrected description for a capability the agent misused.
- Missing doc — a suggested knowledge page for a gap the agent could not answer.
- Eval case — a regression test capturing an input and the behaviour it should produce.
Reviewing a proposal
Each proposal shows why it was drafted and the exact change it would make, in plain fields rather than raw data. You act on it from the detail pane.
- Tabs — Proposed, Accepted, and Dismissed.
- Accept and apply — applies the fix; an accepted eval case is added to your regression dataset.
- Dismiss — discard a proposal you do not want.