docs(handoff): mention-review-tui — the octo replacement, Phase 0+1 built and algorithm-reviewed

[ZacxDev]

Canonical handoff for the new effort: replacing nvim-octo with a purpose-built Go/Bubble Tea PR-review TUI.

Records what this session produced:

• proposal merged (docs(proposal): replace nvim-octo with a purpose-built PR review TUI #1696, 02b54fe4)
• Phase 0's three kill criteria, all passed, with numbers
• Phase 1 shipped as feat(mention-review): Phase 0 + Phase 1 of the PR-review TUI, plus a third gate tier #1698, plus a third gate tier
• the-algorithm review, and the four fixes applied and rebased

🔴 Carries two corrections to claims made earlier in the session, both re-measured:

• "local git first (~0ms)" is wrong on the first read — a PR-ref fetch is 0.84–0.90s, no cheaper than the API. Local wins only on the 2nd..Nth read, so Phase 3 must implement a race.
• only ~8.4% of clickable repos have a local clone (33 of 371), so the API is the majority path, not the fallback.

Closing condition is a judgement: the operator uses the TUI for a real review and says it beats octo at reading a diff. Phase 4 (deleting nvim-octo) is gated on it.

🤖 Generated with Claude Code