PR #879 follow-ups: amend design 007 for the claude-opus-5[1m] pin nf-6qzo ← Beads

closed priority 2 task unassigned activesession:pin-reviewer-model

Twinned from https://github.com/cfsh/ninefold/issues/881 by tools/beads/import_github.py's reconcile pass.

Suggestions from ninefold-reviewer's review of #879.

Amend design 007 so its model string matches what shipped. #879 (007 T1) pins claude-opus-5[1m], not the bare claude-opus-5 the doc specifies: the bare ID selects the 200k-context variant, while every logged review run on the reviewer box resolved to the [1m] 1M-context one (007 §2.2), so the bare pin would have silently narrowed the reviewer's context — 19 of 408 logged passes have read ≥200k tokens. The premise did not survive implementation, which is what design_docs/README.md reserves ## Amendments for.

Stale in design_docs/007-reviewer-verdicts-for-readers.md:

- §0 decision table, D1 row — *"Pin the reviewer's model **explicitly to claude-opus-5**"* - §5 Recommendation, item 1 — *"Pin --model claude-opus-5 in dispatch.py"* - §8 T1 card — Files/Change/Done-when all name claude-opus-5

Shipped: tools/reviewer/dispatch.py:1081-1084.

Per design_docs/README.md: one PR per amendment, separate from the code that motivated it; body left exactly as written, additive ## Amendments entry above the TOC with its own TOC line. Classification (*facts moved* vs *reasoning was wrong* vs *direction changed*) per .claude/commands/design.md step 7 — this looks like *facts moved*: D1's decision stands, the string that expresses it was wrong. Every design_docs/** PR waits for @aedanpope.

Why it is worth doing: a later session implementing T2–T5 reads D1, sees claude-opus-5 in the doc and claude-opus-5[1m] in the code, and "restores" the bare ID.

Refs #879, #872.

Dependencies

None.

Comments

No comments.

Add a comment