v0.1.3

Model catalog refresh from the official command-code@1.24.0.

• COMMAND_CODE_CLI_VERSION bumped from 1.15.1 to 1.24.0 (current official CLI version).
• KNOWN_EFFORTS refreshed against the official command-code@1.24.0 bundled catalog: adds google/gemini-3.7-flash and xai/grok-4.6 reasoning efforts.
• All previously listed models verified unchanged against the official source.

v0.1.2

Fix: network transport failures now surface properly.

• Network-level transport failures now throw LlmError with code TRANSPORT instead of leaking undici's bare TypeError: fetch failed (whose raw detail, e.g. UNKNOWN, previously surfaced as the whole error).
• Both the generate request and a mid-stream read are wrapped, with the original error chained as cause so errorChain renders the full diagnosis.
• Caller aborts propagate unchanged (not relabeled).
• Lets dsh's retry policy recognize and retry transient network failures.

v0.1.1

Quality and robustness improvements.

• Core unit tests (tests/adapter.test.ts, node:test): message conversion, SSE/JSONL stream parsing, catalog parsing, HTTP-error mapping, reasoning-effort exposure.
• fetchImpl injection on the adapter for testability.
• providerRetryPolicy(): declares the default retry policy so metered 429s and transient 5xx are retried at the agent-step boundary.
• HTTP error mapping: 401 → INVALID_CREDENTIAL; 403/429 parse the machine-readable error.code.

v0.1.0

Initial release: Command Code LLM provider plugin for DeepSeek Harness.

• Command Code LLM adapter, ported from pi-commandcode-provider@0.5.1 (MIT).
• commandcode provider route with live model catalog (~54 models) and per-model reasoning-effort metadata.
• Models-page card with an API-key field, backed by the dsh credential seam and the llm-commandcode settings namespace.
• API-key resolution: config → credential reference → environment → official CLI auth file.
• Installable dsh bundle (local path, GitHub, and npm).