pi-commandcode-provider 0.7.11

• Refresh the static capability snapshot to command-code@1.88.1, which publishes metadata for two models the CLI previously did not describe: moonshotai/Kimi-K3.1-Pro finally gains image input, reasoning, and selectable low/high/max efforts instead of the conservative text-only defaults it carried since 0.7.10, and the newly advertised gpt-6.1-terra arrives with image input, reasoning, and low/medium/high/xhigh/max efforts. Neither model publishes a model-specific output limit, so both keep the 65,536-token default. No other model metadata changed (#175).
• Add reviewed display pricing for gpt-6.1-terra, verified against the official pricing page on 2026-10-28: $1 input / $5 output / $0.05 cache read / $1.25 cache write per million tokens up to 272K tokens, and $2 input / $7.5 output / $0.1 cache read / $2.5 cache write above that threshold. The reviewed snapshot now covers all 94 advertised models, so the daily catalog job stops reporting a missing price (#175).
• Stop counting Oh My Pi's skill-description pre-passes as agent requests in test-omp-compat.mjs. OMP 18.13 compresses skill descriptions into routing hints at start-up by calling the configured model once per skill, which added a second POST /provider/v1/chat/completions and failed the request-count assertion against an unmodified main. Pre-passes are now counted separately, keyed on their body shape (user messages only — agent turns always carry a system prompt), and keep the ordinary mock response under the 403 upgrade_required scenario so the fallback stays the only failing path. Test-only change; verified against OMP 18.12.6 and 18.13.2 (#176).

Validation

• Release CI and packed/registry package tests with real Pi and OMP against mock APIs.
• No live Command Code account test is performed by this workflow.
• Commit: f4ab05c
• Release workflow