feat(supergroups): lightweight RCA prototype#110191
Conversation
Add max_iterations field to ExplorerChatRequest TypedDict and SeerExplorerClient. When set, it's passed through to the Seer API to limit agent iterations, enabling lightweight/fast Explorer runs. Co-authored-by: Claude <noreply@anthropic.com>
…er RCA Creates a function to trigger low-cost, fast Explorer RCA runs with max_iterations=3 and intelligence_level=low. Results are stored with category_key=lightweight_rca for later quality evaluation against full RCA. Co-authored-by: Claude <noreply@anthropic.com>
github-actions
bot
added
the
Scope: Backend
Call trigger_lightweight_rca() inside _trigger_autofix_task alongside trigger_autofix_explorer(), so it only fires on the explorer code path for Seer customers. Failures are caught and logged without blocking the main autofix flow. Lightweight runs remain invisible to users via their separate category_key="lightweight_rca". Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
![sentry[bot]](https://avatars.githubusercontent.com/in/12637?s=60&v=4){kind=small}
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Lightweight RCA runs are meant to be invisible, but AutofixOnCompletionHook sends external webhooks (seer.root_cause_completed) and triggers supergroup embeddings on completion. Pass on_completion_hook=None instead to avoid unintended side effects. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
| stopping_point=stopping_point, | ||
| ) | ||
| try: | ||
| trigger_lightweight_rca(group) |
There was a problem hiding this comment.
will the lightweight runs be exposed to the user through the history? not sure if there's a way to trigger an explorer run and keep it hidden to the user
There was a problem hiding this comment.
seems like today in Sentry whenever we get runs we use the "autofix" category so we should be ok. The Seer API technically though could return them if asked without a category, I dont see a place in our product where we ever just expose all Explorer runs, though we might at some point... might be safer to have some way to hide it in Seer as experimental or somehing so it doesnt get accidentally exposed, but it doesnt seem like a must right now
There was a problem hiding this comment.
hm i thought if you open explorer (cmd /) and then do /resume you could see all your runs. not working for me though. agreed that i'm not too concerned about it right now, but might be worth asking the explorer team
There was a problem hiding this comment.
oh interesting was not aware, ill ask them if there are surfraces we are unaware of
There was a problem hiding this comment.
we should be fine it seems, the /resume filters on user_id which autofix runs (and our new ones) dont set so they dont show up. Asked them about it and nothing else came up so I think were good
| project=group.project, | ||
| user=None, | ||
| intelligence_level="low", | ||
| max_iterations=3, |
There was a problem hiding this comment.
re: kush's comment in slack, i'm thinking about it and 3 iterations does seem quite low. maybe restricting the number of tools / making the tool call returns smaller will be a better optimization, but since we probably want to test this out and optimize from there, this isn't blocking
There was a problem hiding this comment.
from Kush's comment this might be blocking, I am looking into it on the Seer side trying to understand, might have something to change there before we merge this
|
|
||
| short_id = group.qualified_short_id or str(group.id) | ||
| title = group.title or "Unknown error" | ||
| culprit = group.culprit or "unknown" |
There was a problem hiding this comment.
just curious, what is culprit on issues? do we pass this to normal explorer?
There was a problem hiding this comment.
yea its the same as the regular one:
3 participants
Summary
Adds a lightweight Explorer RCA that runs in parallel alongside the main explorer-based autofix, for debugging and evaluation purposes.
SeerExplorerClientgainsmax_iterationssupport — allows capping the number of agent iterations per runtrigger_lightweight_rca()— fires a low-intelligence, limited-iteration (3 max) Explorer RCA on an issue, gated by its own feature flag (projects:supergroup-lightweight-rca)_trigger_autofix_task— lightweight RCA triggers only on the explorer code path (alongsidetrigger_autofix_explorer), so it only fires for Seer customers using explorer-based autofix. Failures are caught and logged without blocking the main flow.Why this is invisible to users
Lightweight runs use
category_key="lightweight_rca", while the frontend and API querycategory_key="autofix"— so these runs never appear in the UI or API responses.Feature flags
projects:supergroup-lightweight-rcaorganizations:seer-explorer+organizations:autofix-on-explorerRelates to ID-1385