turn-boundary commitment registry for [Prime Agent](https://github.com/PrimeIntellect-ai/prime-agent).
an extension for [Prime Agent](https://github.com/PrimeIntellect-ai/prime-agent) that stops the agent from forgetting things you asked for.
when you ask for two or three things and the agent branches onto one, the others quietly lose
salience and never come back. this extension keeps an explicit thread registry. a small observer
model runs once per request boundary, updates thread state from evidence, and the still-open
threads are re-injected into model context on every later request.
### the problem
if you ask the agent to do two or three things in one message (like *"fix this bug, add tests, and update the docs"*), it usually starts working on the first thing, goes down a rabbit hole, and completely forgets the rest. even worse, when conversation history gets summarized or compacted, older requests disappear completely.
the observer never closes a thread. it can only propose completion. closure is a host decision.
### how this fixes it
this extension makes sure the agent actually finishes everything you asked:
1. **splits your requests**: breaks compound messages into separate tasks so nothing gets lost.
2. **keeps open tasks visible**: pins still-unfinished tasks into the prompt on every turn, even after compaction.
3. **checks the work**: a small background model watches what gets done and stages tasks for completion when work lands or tests pass.
4. **prevents false claims**: the background model cannot mark tasks "done" on its own. tasks only close when verified by real passing commands or when you accept them.
## requirements
- Prime Agent 0.9.1, or a compatible extension API