Skip to content

cc-explorer: measure conversion-without-use (convert_session created but never resumed) #40

Description

@coryking

Problem

A convert_session fork that is created and then never resumed is waste — we copied a whole transcript and never interrogated it. The reaper (PR #38) sweeps these as garbage, but sweeping them hides the signal. A high rate of pristine (never-resumed) conversions is not a disk-hygiene problem — it's a workflow signal: forks are being created speculatively and abandoned.

Why it's worth surfacing, not just sweeping

If the interrogate flow routinely creates forks it never uses, that says something about how the flow is being driven (over-eager conversion, abandoned interrogations, a step that converts before deciding whether it needs to). Cleanup throws that information away. Measuring it could inform whether the conversion step should be lazier, gated, or prompted differently.

Breadcrumbs (not a prescription)

  • The distinguishing predicate already exists: pristine (not growth_exceeded) vs. grown tells "created but never resumed" from "actually used."
  • cc-explorer already owns the forensics surface (list_session_agents, audit_session_tools); a "conversions created vs. conversions resumed" read would live naturally there.
  • Tension to resolve: the reaper deletes the very artifacts this metric would count, so the signal has to be captured before/independently of reaping (e.g., the reaper could report what it swept).

Deferred from PR #38.

Metadata

Metadata

Assignees

No one assigned

    Labels

    cc-explorerMCP server and typed JSONL toolkitenhancementNew feature or request

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions