Feature request: allow skipping context injection per session #322
Replies: 1 comment
|
Adding another use case that I think argues for a global-disable option alongside the per-session mechanism described here. I work across multiple large repositories (millions of lines of code each), frequently task-switching between development, code review, and design work using git worktrees. For me, context from the previous session is usually noise and wasted tokens — I'm rarely continuing yesterday's thread. The injection fires unconditionally on every SessionStart, and with CLAUDE_MEM_CONTEXT_SESSION_COUNT validated to a minimum of 1 (can't be zeroed out). What I'd find useful is a two-level approach:
This is the inverse of the current model, which is always-on with no escape hatch. For users like me where cross-session continuity is the exception rather than the rule, the default should be opt-in. The per-session skip described in this discussion handles the opposite case (usually-on, sometimes off) — both should exist. |
Uh oh!
There was an error while loading. Please reload this page.
Currently, Claude-Mem automatically injects context from previous sessions at the start of every new session. There is no documented or supported way to start a “fresh” session without prior memory being injected.
This makes it difficult to:
Describe the solution you'd like
Add an explicit opt-out mechanism to skip context injection for a session.
The default behavior should remain unchanged, but users should have a clear and intentional way to start a clean session when needed.
Additional context
The documentation mentions “Context Configuration” and fine-grained control over injected context, but there is currently no visible or documented way to disable injection per session.
Adding this feature would improve user trust, flexibility, and usability—especially for exploratory, sensitive, or debugging workflows—without breaking existing behavior.
All reactions