paradigmxyz/centaur

Closing as we are unlikely to introduce other ways to interact with Centaur outside of natural language back and forth. We can always re-open later if we want to add some common quick actions pattern later.

Release chart CI workflow is now functional. Closing.

The patch in the issue body is enough to land this as a very small workflow PR:

- replace `git switch -` with an explicit `${GITHUB_REF_NAME:-main}` switch after orphan bootstrap
- add the 1Password Helm repo before `helm lint contrib/chart`

If nobody has picked it up yet, I can open the PR from a fork and keep the validation note tied to the failed Release Chart run.

hmm - we already have steering / stopping using `@ stop` - do you think a stop button would be better UX here?

Here is the PR - #174

This also requires a new transform in `iron-proxy` - [that change is also ready](https://github.com/kkennis/iron-proxy/pull/1), and messaged @mslipper via X DMs to see if I could open the upstream PR.

We would love to get this in so we can get the team using Centaur. Cheers!

Maybe it could also be useful to add a Retry button in case a job fails. For example, in the image below, the job failed, but the bot didn’t send any feedback

Happy to! I have a local workaround which passes the smoke test but requires iron-proxy changes. Will look for something more robust over the weekend

Makes sense - wanna do a stab at it? Also @pseudolabel cc re subscriptions

Fresh blocker check on 2026-05-20 22:34 UTC:

- GitHub token scopes are `admin:public_key, gist, read:org, repo`; it still lacks `workflow`, so I cannot push the Release Chart workflow fix from this sandbox.
- `call tools` still returns `401 Invalid or expired sandbox token`, so I cannot use Centaur API/vlogs/workflow/Slack tooling to verify live Slackbot rollout or run a real Slack smoke.
- No follow-up PR matching the Release Chart/chart-releaser blocker is present; issue remains the handoff for deploy-capable verification.

Remaining verification before closing the delivery-flakiness task: confirm fixed Slackbot is live, run a real Slack-thread smoke, and check clean post-deploy logs.

Manual chart publish completed from the sandbox as a workaround for the failed Release Chart workflow.

Evidence:
- `gh-pages` updated to commit `068587e390b13fd80b7a82d4c948248276a1a174` with `centaur-0.1.38.tgz` and `index.yaml`.
- GitHub Pages build `1007661260` completed with status `built` at `2026-05-20T19:46:50Z`.
- `helm repo add centaur-manual https://paradigmxyz.github.io/centaur && helm search repo centaur-manual/centaur --versions` now returns chart version `0.1.38`.

Remaining blocker: the live cluster has not rolled Slackbot yet. Logs after `2026-05-20T19:46:50Z` still show Slackbot pod hash `66967fdd97` (`centaur-centaur-slackbot-66967fdd97-pzljx`), and fresh API `workflow_history_backfill_skipped` / `IDEMPOTENCY_PAYLOAD_MISMATCH` warnings continue from pod hash `79cdfc8f5b`.

Still needed: run the deployment/Helm upgrade or restart path against chart `0.1.38`, then verify a real Slack-thread turn and clean logs for `final_delivery_failed`, `final_delivery_dead_lettered`...