Summary
Long assistant replies are truncated mid-sentence in the Slack stream, cutting off the completion summary at an arbitrary character boundary. The user receives a partial message with no indication that the content was cut.
Observed behavior
- Assistant completes a large multi-file code change and begins posting a summary.
- The summary message is cut off mid-sentence:
...confirmed from prior session's sentry/snuba/utils.py research) 2. Have its own URL builder pointing to... — body ends abruptly.
- No truncation marker, continuation message, or ellipsis is shown.
Root cause (hypothesis, medium confidence)
Likely a Slack message character limit (max ~4000 chars for a single message) hit without a graceful continuation strategy. The streaming path may not chunk long replies into multiple messages or canvas artifacts when the limit is approached.
Potential fixes
- When a streaming reply exceeds a safe character threshold, automatically overflow into a Slack canvas or a follow-up thread message.
- At minimum, append a truncation notice ("response was cut off — see full output in [canvas link]") rather than silently ending.
Related
Action taken on behalf of David Cramer.
Summary
Long assistant replies are truncated mid-sentence in the Slack stream, cutting off the completion summary at an arbitrary character boundary. The user receives a partial message with no indication that the content was cut.
Observed behavior
...confirmed from prior session's sentry/snuba/utils.py research) 2. Have its own URL builder pointing to...— body ends abruptly.Root cause (hypothesis, medium confidence)
Likely a Slack message character limit (max ~4000 chars for a single message) hit without a graceful continuation strategy. The streaming path may not chunk long replies into multiple messages or canvas artifacts when the limit is approached.
Potential fixes
Related
Action taken on behalf of David Cramer.