[igorcosta]

Weka is removed from the CLI catalogue while its direct /v1/decisions compaction route remains intact. Explicit Weka selection says: "Weka is available through the Autohand API, the Agent SDKs and the Console, not in the CLI. Pick another model with /model or --model." The provider guide uses Igor's corrected API, Agent SDK and Console wording and names the decisions endpoint.

The catalogue also adds Sonnet/Opus/Haiku 5.5 and Fable 5.1 to Anthropic and OpenRouter, with Sonnet 5.5 first and default. Accepted-plan auto approval permits file-edit tools only; shell operations still ask. Manual acceptance overrides YOLO, auto-confirm and saved approvals. Execution and permission overrides end after success, failure or cancellation. Rolling-window errors show trusted reset-credit billing links; monthly behavior is unchanged.

Validation of the Console wording follow-up at 3e69631ee8a3abe0d6785b799b69fece2a7db82b:

• All 40 focused provider/catalogue tests pass locally using the existing dependencies. Existing terminal-test expectations are updated; the built-terminal check runs in hosted CI.
• The catalogue is byte-for-byte unchanged by this follow-up and still contains zero Weka occurrences. The exact requested guard and provider-guide strings were checked.
• The normal hosted matrix and Full proof run for this head; their results remain pending until completed. No heavy local build or dependency installation ran while the Mac is below its disk floor.

Previous full validation at parent c8f30e04437a7ac89bddecabdab2f88537f90d85:

• Complete proof, including lint, typechecking, 12,807 unit tests and 428 terminal tests, passed. Existing skips: 61 unit and 6 terminal.
• PR matrix and push matrix passed, including all three platform builds. Original failed attempts are retained.
• Released-client picker checks on the unchanged candidate catalogue confirmed no Weka on 0.9.8 and Sonnet 5.5 first on 0.9.9 using isolated homes. No provider model request was made.

Copy follows Igor's 29 October correction relayed by Nikita. The other reviewed Weka and reset-credit requirements remain in place. Production reset-credit binding presence was last successfully checked at 28 October 13:30 UTC; it must be rechecked before any inference deployment, with no credential or binding value exposed.

Merge remains held for the required GitHub review and explicit go. The Main ruleset requires an approving PR review and linear history: https://github.com/autohandai/code-cli/rules/13813163. No rules or workflow settings were changed.

Once cleared, the approved route is a squash merge containing [skip ci], followed by dispatching only publish-model-catalog.yml on main. This preserves linear history and publishes the approved catalogue without an unapproved alpha/npm release. Runtime changes ship with 0.9.10. Catalogue publication and post-publication fresh-home checks remain pending. No merge, catalogue publication, npm release or deployment was performed for this follow-up.

Tier-policy test correction

The test (2) job in run 40768125293 had exactly one failed assertion: its expected Weka text omitted Console. Updated that expectation and the test name. No production code changed. Reproduced the single failure before the edit; all 58 tier-policy, provider and catalogue-policy tests now pass locally. Hosted CI and complete proof are pending on this head.

No merge, release, catalogue publication or deployment performed.