Changes:
- Updated the request to the OpenAI API in `receive-from-github.js` to
send a request to the `chat/completions/` API and updated the request
body to be compatible with the `/chat/` API.
(https://platform.openai.com/docs/api-reference/chat/create)
Upgrade to GPT-4. Increase max tokens 4x up to 8000 (aka the maximum
character length of issue description that it can pontificate about in
the form of a short haiku)
.
Changes:
- Updated `receive-from-github.js` to only send requests to freeze and
unfreeze the Fleet repo when the PR it is checking is from the
fleetdm/fleet repo.
Changes:
- Updated the `receive-from-github` webhook to tolerate failed requests
to the Merge Freeze API. If a request fails, the webhook logs a warning
and continues as if the main branch is not frozen.
---------
Co-authored-by: Mike McNeil <[email protected]>
Changes:
- Updated the conditional statement that determines if we will be
notified in Slack to filter out commit comments and deleted comments
from Fleet team members with uppercase letters in their GitHub
usernames.
Improve https://github.com/fleetdm/fleet/pull/9336 providing default
settings similar to the OpenAI playground, including a higher
temperature and number of maximum tokens. Also fixes prefix trimming.
This improves https://github.com/fleetdm/fleet/pull/9336 by eliminating
junk text and encouraging better replies. Uses an h1 to emphasize the
issue title, so that short issues don't get weird and truncated, and the
bot reply stays focused on the main point of the issue.
* add #handbook label to handbook PRs
* Update get-is-pr-only-handbook-changes.js
* remove while loop from _.all()
* revert changes to github token name
* Simplifications
Co-authored-by: Mike McNeil <[email protected]>
A timeboxed workaround for an intermittent issue where the mergefreeze API reports that a repo is frozen: false, when the mergefreeze UI says that it is actually truly frozen.
Turns out the issue being debugged in https://github.com/fleetdm/fleet/pull/6906 was in the API itself-- somehow it can get stuck on frozen: false. If that happens, the solution is to freeze/unfreeze everything in the dashboard
* Tool: Bring back mergefreeze API integration
Context: https://github.com/fleetdm/fleet/pull/5628#issuecomment-1196175485
Unfortunately, the API doesn't work.
* Lay out how this would work in the database - but instead, do it ephemerally for now
* Remove model
* Maintain state (the easy way for now)
* trivial (comments)
* Remove old unfreeze/freeze logic
* trivial (clarify comment)
* Trivial (fix weird character)
* Extrapolate DRI mappings into config.
* Explain why this exists
* Extrapolate logic
* Use extrapolated logic + add 5 second wait time to prevent accidents + clean up
* Use extrapolated logic and fix omission in helper
* Make freezing actually happen and document usage
* In script, don't freeze PRs as long as they're preapproved to be edited by SOMEBODY
* Lint fixes
