https://docs.all-hands.dev/ logo
Join Slack
Powered by
<@U0A9RLDPQU8> Can we please disable the automatio...
# automations-debug
c
@modern-fireman-99755 Can we please disable the automations that were inherited from stale.yml? They post a comment mentioning maintainers from a maintainers file that is itself stale, and I’m cleaning stuff up.😅 I recognize I just got a notification too many, and maybe got this in the wrong order (I’ll come back with other proposals). Either way, we have better ways today to handle things, and we can iterate on those.
m
Sorry i don’t get it :)
c
For example, this comment doesn’t seem to come from the workflow I removed: https://github.com/OpenHands/software-agent-sdk/issues/4740#issuecomment-6083508962 It comes from an automation inside the VM, I assume? I’d like these comments to not happen anymore
Or maybe it was coming from the GH workflow? The agent tells me timing doesn’t match lolsob
m
Ah yes, I think there are still leftover workflows 🙂
or two different workflow to be precise
But the nudge workflow is now on the VM too
c
Yes, it can be disabled/removed from the VM, right?
m
@fresh-machine-37269 Can you check if the nudge workflow for stale issue is present on the VM and in the repo?
👀 1
f
🤖 On it! View progress here: oss-agent-canvas.ngrok.dev/conversations/…
✅ Done! Here's what I found: In the repo — yes, and there are actually two of them. They live in 
OpenHands/oss-agent-canvas
(the repo that stores the Canvas automations), under `automations/`: • 
stale-issue-nudge-multi-repo-openhands-software-agent-sdk
→ enabled (
enabled: true
). Cron 
0 15 * * 2,5
(Tue & Fri 15:00 UTC). Covers 
extensions
, 
OpenHands
, 
software-agent-sdk
, 
automation
, and `enterprise`; it nudges all of them and also auto-closes inactive issues on 
software-agent-sdk
(tagging the 
MAINTAINERS
file). This is almost certainly the one that posted the comment on issue #4740 — its output matches the "[Automatic Post]: This issue has had no activity for N days…" format exactly. • 
stale-issue-nudge-software-agent-sdk
→ present but disabled (
enabled: false
). Looks like the leftover single-repo version. There's no 
stale.yml
(or any nudge/stale) GitHub Actions workflow left in 
software-agent-sdk
or the other repos — those are gone. On the VM — I don't see any nudge workflow installed on this machine: no automation definitions, no 
automation.yaml
, no cron job, no systemd unit, and no process running the nudge script. The nudge runs as a Canvas automation from the backend (the repo is literally synced from that server — "Sync automations from agent server" commits), not from this VM's own filesystem. (I couldn't query the backend live — its ngrok endpoint is currently offline — but since the multi-repo entry is 
enabled: true
, it is deployed and active.) So, short version: the nudge workflow is not on the VM itself, but it is in the repo — twice, and the multi-repo one is the live one. To stop it, you'd disable/remove the multi-repo automation on the Canvas backend (flipping 
enabled: false
and syncing), same as was done for the single-repo one.
m
@fresh-machine-37269 Can you confirm that we do have a workflow in the repo to nudge issues? And that we do have it twice? Can you point me to the files and line numbers?
👀 1
f
✅ Done! Confirmed. Both live in 
OpenHands/oss-agent-canvas
 (private repo, 
main
branch), under 
automations/
. There is no GitHub Actions workflow for this — these are Canvas automation bundles, each a directory with an 
automation.yaml
(the definition) plus a 
tarball/main.py
(the script). 1. Single-repo version (disabled) 
automations/stale-issue-nudge-software-agent-sdk/automation.yaml
• line 1: 
enabled: false
← switched off • line 5: 
name: Stale Issue Nudge -- software-agent-sdk
• lines 13–16: cron trigger 
0 15 * * 2,5
(Tue & Fri 15:00 UTC) 
automations/stale-issue-nudge-software-agent-sdk/tarball/main.py
• line 36: 
REPOSITORY = "OpenHands/software-agent-sdk"
• line 184: 
def build_nudge_comment(...)
(the "[Automatic Post]…" text) • lines 250–255: the 
POST /repos/.../issues/{n}/comments
that posts the nudge 2. Multi-repo version (enabled — this is the live one) 
automations/stale-issue-nudge-multi-repo-openhands-software-agent-sdk/automation.yaml
• line 1: 
enabled: true
• lines 5–6: 
name: 'Stale Issue Nudge (multi-repo): extensions, OpenHands, software-agent-sdk, automation, enterprise'
• lines 14–17: same cron 
0 15 * * 2,5
automations/stale-issue-nudge-multi-repo-openhands-software-agent-sdk/tarball/main.py
• lines 43–71: the 
REPOS
list — 
extensions
, 
OpenHands
, 
software-agent-sdk
(lines 55–59, with 
close_enabled: True
+ 
mention_maintainers: True
), 
automation
, 
enterprise
• line 217: 
def build_nudge_comment(...)
• line 248: 
def close_issue(...)
• line 300: the 
POST .../issues/{n}/comments
that posts the nudge So yes — two definitions, same schedule, both nudging issues; the multi-repo one is the only enabled one, and it's the one that produced that comment (it's also the one that auto-closes stale 
software-agent-sdk
issues after 30 days of genuine inactivity).
m
Ok but I'm speaking the OpenHands/software-agent-sdk do we have workflow for that?
@fresh-machine-37269 Ok but I'm speaking the OpenHands/software-agent-sdk do we have workflow for that?
👀 1
f
✅ Done! No — not in 
OpenHands/software-agent-sdk
itself. There's currently no workflow in that repo that nudges stale issues. Here's the precise picture: It used to have one, and it's now gone. 
.github/workflows/stale.yml
("Close stale issues", using 
actions/stale
— mark stale after 40 days, close after 10 more) was deleted on 2026-10-09 by commit 
dd1af943
— "chore: remove stale triage workflow and maintainer roster (#5683)". That's the cleanup you did; before deletion it held 
stale-issue-message
/ 
close-issue-message
and 
days-before-stale: 40
/ 
days-before-close: 10
. Nothing replaced it inside the repo. I went through every file in 
.github/workflows/
on the current default branch — none of them nudge or comment on stale issues. The only "stale" hits are unrelated comments in 
version-bump-prs.yml
(lines 141, 367, 369, 525, 817). A repo-wide code search for 
nudge
only turns up the SDK's own agent stuck-detector nudge (
openhands-sdk/openhands/sdk/conversation/stuck_detector.py
etc.), which is a completely different thing — nothing to do with issue triage. So the nudge that's hitting the SDK repo is the external one. The 
OpenHands/software-agent-sdk
entry in the multi-repo Canvas automation (
OpenHands/oss-agent-canvas
, 
automations/stale-issue-nudge-multi-repo-openhands-software-agent-sdk/tarball/main.py
, lines 55–59, with 
close_enabled: True
, 
mention_maintainers: True
) is what operates on the SDK repo. It's a Canvas automation, not a GitHub Actions workflow in the SDK repo — which is exactly why the timing didn't line up with the workflow you removed. To stop the comments on SDK issues, you disable that Canvas automation (the single-repo one at 
automations/stale-issue-nudge-software-agent-sdk/
is already 
enabled: false
).
c
@modern-fireman-99755 I had just removed it: https://github.com/OpenHands/software-agent-sdk/pull/5683/changes
But this wasn’t responsible for the comment, as the bot says; the automation-style stale is
m
ah ok I will disable that... sorry I'm confused
❤️ 1
thanks on a stick 1
Open in Slack
PreviousNext
Powered by