Bug
In 0.4.7, codealmanac sync can ingest transcripts produced by CodeAlmanac's own lifecycle workers. The run ledger records the worker provider session id in RunRecord.harness_transcript.session_id, but sync does not currently use that state when selecting transcript candidates.
Impact
If CodeAlmanac runs an ingest/build/garden operation through Claude or Codex, that provider transcript can later be rediscovered by sync and queued as source material for another ingest. This creates self-ingestion noise and can cause the wiki to document CodeAlmanac's maintenance work rather than user project work.
Expected
Sync should skip discovered transcript candidates whose (provider, session_id) matches a harness transcript already recorded for the same repository.
Actual
Sync filters discovered transcripts by registered cwd and active window, but does not exclude sessions already owned by CodeAlmanac runs.
Current affected surface
Current source has two worker harnesses and two transcript discovery apps:
No ChatGPT/Gemini/other worker transcript adapter exists today.
Proposed fix
Have sync read recorded harness transcript refs from the run ledger and skip matching transcript candidates before grouping eligible transcripts for ingest.
Bug
In 0.4.7,
codealmanac synccan ingest transcripts produced by CodeAlmanac's own lifecycle workers. The run ledger records the worker provider session id inRunRecord.harness_transcript.session_id, but sync does not currently use that state when selecting transcript candidates.Impact
If CodeAlmanac runs an ingest/build/garden operation through Claude or Codex, that provider transcript can later be rediscovered by sync and queued as source material for another ingest. This creates self-ingestion noise and can cause the wiki to document CodeAlmanac's maintenance work rather than user project work.
Expected
Sync should skip discovered transcript candidates whose
(provider, session_id)matches a harness transcript already recorded for the same repository.Actual
Sync filters discovered transcripts by registered
cwdand active window, but does not exclude sessions already owned by CodeAlmanac runs.Current affected surface
Current source has two worker harnesses and two transcript discovery apps:
No ChatGPT/Gemini/other worker transcript adapter exists today.
Proposed fix
Have sync read recorded harness transcript refs from the run ledger and skip matching transcript candidates before grouping eligible transcripts for ingest.