Note on repo2graph Hive interaction

[pitoi]

Notes on this run
https://jobs.stakwork.com/admin/projects/103071021

1. No spinner when a button artefact is pressed, so the user can tap multiple times
   After selection:

• replace button tapped with spinner
• grey out unselected artefacts

2. The "value" is displayed when the user taps on a button artefact instead of the "Label"

3. Workflow issue: change wording so it doesn't sound like we're asking the same question twice and displaying the page twice:

4. Workflow wording:
   We're really asking if the user wants to proceed with generating a strategy

5. feature for hive
   Render the page when possible.
   We could create new "artefacts" that are appropriate for each type of user feedback requested.
   IE Color picker
   web page confirmation: render page

6. Workflow issue
   https://jobs.stakwork.com/admin/workflows/43926/edit
   Gather feature context.... JSON was 40k lines long with code and map

7. Workflow issue: strategy step
   Here is correctly

8. Hive/Workflow: Approve
   Typing the word "APPROVE" Is required to break out of the while loop and generate the code.
   Could be replaced by an artefact that says "Create code"

9. Workflow issue: Max tokens hit
   We tried to send too much context to Claude

[Evanfeenstra]

Tiktokenizer should be used to approximate token length, and > 100k for context (from the code_bodies endpoint in 2B) needs to filtered down

[Edvardoh]

On #8 - I changed it to an APPROVE button instead of having to type out the word APPROVE, mostly to get the hang of the artifacts object. Can change the text to "Create Code" instead if we want, but the button makes more sense here

[humansinstitute]

1. We have a bounty going in for button deactivations (also solves sending action message back) i just need it rebased to merge

2. The item displayed is actually the "message" we send to the chat webhook.

i.e. we create a message with the value from the button. This is created as a mesage and added to chat history so we have a record of the actual conversation. It also ensure that when we return to a page we can render the chat history.

Whilst this could be updated we'd be layering in a bunch extra logic into "some" chat bubbles. Is this a hard requirement or could we keep values and labels close so we can keep rendering the actual message sent to the AI?