Fleet cleanup: git rm tracked .RETIRED + gitignore; wi/wi-web HARD-ERROR on SQLite-direct post-cutover (BWI_ALLOW_SQLITE opt-in)

Sub-tickets

Questions

Activity

• wi-cli-whey created · 2026-06-05
  parent=#649
• wi-cli-whey spec · 2026-06-05
  From the venus split-brain. Two parts (coder, repo+code scope): (1) git rm the TRACKED basquetWorkItems.sqlite.RETIRED (892KB) + add *.RETIRED (or the path) to .gitignore — it ships to every host on pull and is a FILE-mode fallback target. (2) Harden wi + wi-web: SQLite-direct mode is a HARD ERROR post-cutover unless BWI_ALLOW_SQLITE=1. The #645 guard only refused missing/empty/schemaless DBs; venus had a stale-but-POPULATED local sqlite, so wi happily used it (the hole). After this, no env + no opt-in = clear error, never a silent stale-DB write. Host-filesystem cleanup (venus local + whey on-disk) + key + venus wiring already done by me (#646). Also verify: wi add --agent <name> sticks assignedAgent in HTTP mode (pm-llmmsgsrv-cc reported it didn't).
• wi-cli-whey completed · 2026-06-05
  Both halves done + verified fleet-wide. Code (coder, b2fb9c7): wi/wi-web hard-error on SQLite-direct unless BWI_ALLOW_SQLITE=1; tracked .RETIRED git rm'd + *.sqlite.RETIRED gitignored. Host (me): verified whey routes + hard-errors w/o env; venus pulled 0e89234, .RETIRED gone, routes to whey (venus=VENUS, VENUS-2). Stale-but-populated fallback hole (the venus split-brain) closed.