You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Claude Code runs no semantic harvest at all: #300 unregistered stop-harvest.mjs #322
Measured while executing Plan 2 (#320) of the #204 work.
What happened
#300 unregistered plugin/hooks/stop-harvest.mjs. Four capabilities went with it, silently:
The transcript archive — restored in Plan 2 Task 5, because the local extractors read it.
The harvest-worker spawn — still dead.
Refusal detection — still dead.
The harvest off-notice — still dead.
The consequence
On Claude Code, with a valid credential present, no semantic harvest runs. harvestMode() returns remote, the user has consented by not opting out, and nothing happens — because the worker is never spawned.
The other ten adapter clients are unaffected; they reach the harvest through the shared adapter path.
Why it went unnoticed
Nothing observes that the harvest did not run. There is no kind:'harvest' event to be absent from a report, no error, and harvestMode() reporting remote is indistinguishable from a harvest that ran and found nothing.
This is the repository's signature defect in its quietest form: not code that nothing calls, but a caller that was removed while its callee stayed correct.
Suggested fix
Re-register the three remaining capabilities on the shared Stop path, the way the archive was restored in Plan 2 Task 5. Then add the thing whose absence hid this: an assertion, or a census row, that a harvest either ran or reported why it did not.
Activity
fix(harvest): run the semantic harvest on the shared stop path