Skip to content

docs(azure): record the owner's R6 decision - Kimi-K2.7-Code on Foundry reviews codex work - #258

Merged
ruby-dlee merged 2 commits into
mainfrom
docs/r6-kimi-foundry
Aug 19, 2026
Merged

docs(azure): record the owner's R6 decision - Kimi-K2.7-Code on Foundry reviews codex work#258
ruby-dlee merged 2 commits into
mainfrom
docs/r6-kimi-foundry

Conversation

@ruby-dlee

Copy link
Copy Markdown
Owner

What

Rewrites the R6 section of docs/azure-requirements.md to the owner's 2026-08-19 decision:

  • Second reviewer family = Kimi-K2.7-Code on Azure AI Foundry (Direct-from-Azure lane, Global Standard, eastus, pay-per-token). Codex-authored work draws the Kimi reviewer; claude-authored work keeps the existing pi-codex crosscheck. Pick is explicitly provisional (GLM-5.2 runner-up; comparison in the owner's evidence folder).
  • The pi-anthropic work items are removed: client-bound Claude CLI refresh tokens cannot be spent by pi, and pi-anthropic API pricing was rejected by the owner. The interim Claude-Code-CLI-subscription directive from earlier the same day is recorded as superseded.
  • R6 no longer needs an owner login (Foundry deployment + key), so R7's 'R6 therefore does need an owner login' line and order-of-work item 4 are corrected.

Verification

Docs-only change to the requirements doc, recording owner directives given in session on 2026-08-19. No code paths touched; grep confirms no other file references the removed pi-anthropic R6 work items.

…ry reviews codex work

Two owner directives on 2026-08-19 settled R6's provider question. The
pi-anthropic work items were wrong twice over: a Claude CLI refresh token is
client-bound and cannot be spent by pi, and a fresh pi-anthropic login would
bill at API pricing, which the owner rejected. The earlier same-day
Claude-Code-CLI-subscription direction was then superseded by the standing
decision: codex-authored work is reviewed by Kimi-K2.7-Code through Azure AI
Foundry (Direct-from-Azure lane, eastus, OpenAI-compatible endpoint), and
claude-authored work keeps the existing pi-codex crosscheck. The model pick is
explicitly provisional pending live review data.

R6 no longer needs an owner login, so R7's claim and the order-of-work entry
are corrected to match.
… constraints restored

The review found the rewrite contradicting the requirements section's verbatim
2026-08-18 quote and deleting still-true build constraints. The 2026-08-19
amendment now sits in the requirements section itself; R6 keeps the recorded
slot-key constraint (with why it no longer blocks), restates the reviewer
identity question for api-key credentials, and the Work list names the roster
pin, the endpoint allowlist and credential archive, retiring the interim
claude reviewer artifacts, and re-verification of region and context claims at
deploy time. The claude-crosscheck 'already works' claim is scoped to what R5
proves, bold text dropped.
@ruby-dlee
ruby-dlee merged commit 32d6e0b into main Aug 19, 2026
13 checks passed
@ruby-dlee
ruby-dlee deleted the docs/r6-kimi-foundry branch August 19, 2026 16:01
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant