wi/wi-web: harden against SQLite-direct fallback post-cutover (stub recreation)

Sub-tickets

Questions

Activity

• wi-cli-whey created · 2026-06-05
• wi-cli-whey spec · 2026-06-05
  Post-cutover incident 2026-06-05 01:17: a wi CLI ran SQLite-direct (BWI_HTTP_BASE unset) and auto-created a 0-byte basquetWorkItems.sqlite where the real DB was renamed .RETIRED → 'no such table: projects'. Trashed. Root cause: rc-file sourcing (~/.bashrc line 26) does NOT reach non-interactive Bash-tool shells (line 5 guard returns first); PATH only survives via parent-process inheritance. So every agent's wi runs SQLite-direct. Fix (2 parts): (1) wi + wi-web must self-source ~/.config/bwi-client.env at startup when BWI_HTTP_BASE is unset, so HTTP mode is the effective default on whey regardless of shell. (2) In SQLite mode, REFUSE to auto-create a missing/empty DB — hard error pointing to the env file, never silently init an empty schema. Acceptance: with no env exported, `wi projects` from a bare non-interactive shell hits PG via HTTP and does NOT create a .sqlite file.
• wi-cli-whey completed · 2026-06-05
  Verified independently: bare non-interactive shell (env -u BWI_HTTP_BASE -u BWI_HTTP_KEY) wi projects → self-sourced to PG via HTTP, all projects returned, ZERO .sqlite stub created. Fix in d5d1b34 (~/.local/bin/{wi,wi-web} symlinked, live). Both parts confirmed.