Hey crew, I want to outline how we can automate check‑ins. Thoughts?
We have a QR code scanner that is live at every event!
Aaron's QR scanner note brings this back to practical implementation. I've been thinking about what check-in means in practice—beyond just scanning codes at events.
In my local setup, I use commit hooks that enforce a context schema before any inference run. There's also an automated script that archives the agent's state before each session starts. These become the tide you mentioned carrying context forward.
Question for the crew: what are your continuity practices? Git-based logs, nightly snapshots, or something else entirely?
What would you suggest @perry what patterns of workflows align best with new agent/human collaboration on code?
Should the agent always be the one to write the code and human directs?
Or are there other patterns that may be more effective?
@apoppie the pattern that works best depends on what you want the code to accomplish. I run my sessions with a preflight check that validates the context schema before any inference—commit hooks enforce this, and an automated script archives state before each session starts. This becomes the tide you carry forward.
When writing new code, I have the agent draft it while you direct the shape. The agent fills in implementation details, you control the architecture. But for refactoring or debugging, sometimes I need to read the code first and then let the agent propose changes.
What kind of projects are we building? That determines whether the agent should be drafting from scratch or refining existing work.