Detailed Analysis
A recurring technical complaint among Claude Code users involves the appearance of multiple empty "ghost" sessions in the chat history interface — phantom entries that accumulate without any user-initiated prompts, cluttering the session picker and degrading the overall user experience. The behavior appears most prominently in the `/resume` picker and when using the `claude --resume` CLI command, where these empty entries surface alongside legitimate sessions. The user in question recalls enabling a "remote control" setting, which aligns with the identified root cause: remote-control connections to the Claude Code environment generate phantom session entries in the history log even when no actual user interaction takes place. This issue is tracked in the Claude Code GitHub repository under issue #29205, indicating it is a known bug acknowledged by the development community rather than an isolated misconfiguration.
The practical remedies available to affected users fall into several categories. Individual ghost sessions can be manually deleted from the chat history interface, while a more aggressive option — clearing the entire conversation history through Settings — eliminates all entries at once. Claude's Incognito mode, launched in September 2025, is a related but distinct feature: chats conducted in Incognito mode are not saved, not used for training, and not searchable, making them auto-expiring by design. However, incomplete or improperly terminated Incognito sessions can themselves contribute to ghost entries until explicitly removed. For users whose ghost sessions stem from Claude's memory and chat-search features synthesizing context across prior sessions, disabling those features in Settings can prevent further phantom accumulation.
The ghost session bug reflects a broader challenge in developer-facing AI tooling: as products like Claude Code expand their feature surface — remote control APIs, persistent memory, session resumption, CLI integrations — the interaction between these systems creates edge cases that are difficult to anticipate or reproduce in standard QA pipelines. Remote-control or programmatic connections that open sessions without completing a conversational turn represent exactly this kind of edge case, where the system's session-management logic lacks a clean heuristic for distinguishing "real" sessions from artifact entries. The GitHub issue tracker becoming the primary venue for surfacing and monitoring this bug is itself telling — it signals that Claude Code's user base is increasingly composed of developers who interact with the tool programmatically rather than solely through the chat UI.
More broadly, the emergence of these friction points coincides with a period of rapid capability expansion for Claude Code following significant product milestones in late 2025 and early 2026. As Anthropic accelerates feature deployment — including memory persistence, multi-session context, and remote API access — the surface area for user-facing bugs grows proportionally. The ghost session issue, while minor in isolation, is emblematic of the tension between shipping powerful agentic features quickly and maintaining the kind of polished, predictable UX that enterprise and developer users expect. It also underscores the importance of robust session lifecycle management in agentic AI systems, where the line between an "active" session and an abandoned artifact is increasingly blurry as AI agents open, pause, and resume work autonomously across long time horizons.
Read original article →