[Requests and ideas] The agent-merge skill is pretty costly when left alone if waiting for human a...

**Mood:** 😊
**Category:** Requests and ideas

The agent-merge skill is pretty costly when left alone if waiting for human approval etc. because the check-ins rely on the agent to fully load the skill file, then perform the script checks only to be told "no changes".

Can the agent merge functionality be made to be far more determinstic and code-driven?

Code is free for polling and finding the changes BEFORE passing off to the merge agent. If nothing to do, the agent should not be invovced at all. Even if there are things to do, by running the context gathering before passing to the agent it can be given a more succint prompt/skill plus direct context instead of asking it to also load said context and how to do it.

With the upcoming token-based pricing this is critical to keep the functionality without wasting millions of tokens (as found by using `/chronicle cost-tips`)

---
| Field | Value |
| --- | --- |
| App version | 0.2.11 |
| OS | Windows 10.0.26200 |
| Theme | GitHub |
| Path | /chat |
| Tenure | Week 3 |