Skip to content

fix(aidd-telemetry): a session run in a git worktree is missing from every report outside that worktree, and lost when the worktree is removed #932

Description

@VincentMenard76

Description

A session run in a git linked worktree is missing from aidd telemetry report run anywhere else, and is lost for good once the worktree is removed.

The hook writes the run journal under the worktree's own aidd_docs/runs/ (git rev-parse --show-toplevel). The CLI reads only the journal of the checkout it runs in, and the sink is filled only from that journal. aidd telemetry on git-ignores aidd_docs/runs/, so git worktree remove deletes the journal with the worktree.

Affected file(s)

  • cli/src/kernel/paths.ts (resolvedRunsDir)
  • cli/src/kernel/reading/repository-root.ts (repositoryRootAbove stops at the worktree's .git file)
  • plugins/aidd-telemetry/hooks/lib/repo.cjs (journal anchored at --show-toplevel)

Expected behaviour

A session run in any worktree of a repository appears in a report run from any checkout of that repository, and still appears after that worktree is removed.

Observed behaviour

One repository, three active worktrees, one Claude Code session each:

Session session_start in its worktree journal Records in the sink
A yes, with worktree_id 0
B yes, with worktree_id 695 (a report was run from B)
C yes, with worktree_id 0

A report from the main checkout sees none of them. Removing the worktree of A or C now leaves no journal to recover their cost from; the transcripts remain but nothing names them.

Proposed solution

  1. Decide where a worktree's journal lives, or how a report gathers every worktree's journal, and write the decision down. fix(framework): a git worktree gets its own journal, by accident rather than by decision #693 raised this choice; feat(framework): measurement that works on five tools, each proven by a session that ran #706 closed it by delivering the worktree_id field (feat(framework): a session says which worktree it ran in #695), and the per-worktree default was kept without a recorded decision.
  2. Whatever is chosen, a journal must outlive git worktree remove.
  3. A report run from any checkout of a repository counts the sessions of all its worktrees. Only the cumulative figure is reported: worktree_id stays in the journal, and the report adds no worktree axis (the task, backlog, flow and agent axes already answer what a piece of work cost).

Context / Technical constraints

  • fix(framework): a git worktree gets its own journal, by accident rather than by decision #693 assumed "sessions from worktree B still contribute their tokens, the sink pools everything, they read as unattributed". That no longer holds: the sink is filled only through the current checkout's journal (read-local-cost-use-case.ts, report-cost-use-case.ts catch-up), so those sessions are absent, not unattributed.
  • Committing journals is not an option: aidd telemetry on git-ignores them on purpose (telemetry-on-use-case.ts).
  • Agent runners (Orca, Claude Code under .claude/worktrees/) give each agent its own worktree, so this is the common case, not an edge case.

Environment

AI tool Claude Code 2.1.283
aidd-cli version 5.4.0
OS Windows 11 Pro (10.0.26100)
node 24.12.0
aidd-telemetry 0.2.0

QA

  1. In a repository with telemetry on, create two worktrees and run one session in each.
  2. From the main checkout, run aidd telemetry report: both sessions appear.
  3. git worktree remove one worktree, run the report again: its session still appears.

Relations

Field Value
parent #631
related #693, #695, #657, #707

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    Fields

    Priority

    None yet

    Projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions