Skip to content

feat: deterministic session labeling from working context (no LLM call required) #246

@RyanLHicks

Description

@RyanLHicks

Problem

Session names currently show only timestamp, message count, and last session messages. When sessions accumulate, users cannot identify which session belongs to which project or goal without opening each one. Creates friction for the user and presents confusion between similar projects.

Prior Pulls

PRs #49 and #51 proposed AI-powered session naming. This proposal takes a different approach —no LLM call, no token cost, no latency.

Proposed Approach

Extract session identity from data already available:

  • Working directory at session start
  • First user prompt (truncated to ~60 chars)
  • Most frequently accessed file or folder
  • Active agent name

Display format example:
[dashboard-project] Fix column mapping | 47 msgs

instead of:
auto_session_20260315_143022 | 47 msgs

Session Grouping

Sessions sharing the same working directory or file pattern could be visually grouped in the /resume picker — without merging or modifying any session data.

Notes

Purely additive. No changes to session storage format. No LLM calls. Backward compatible with existing sessions.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions