feat: add MiniMax CLI harness (chat + TTS)

[octo-patch]

Summary

• Add minimax/agent-harness/ — a new CLI harness for the MiniMax AI platform
• Implements chat via MiniMax's OpenAI-compatible API (models: MiniMax-M2.7, MiniMax-M2.7-highspeed)
• Implements TTS via MiniMax's t2a_v2 SSE streaming endpoint (models: speech-2.8-hd, speech-2.8-turbo), with hex-encoded audio decoding
• Uses MINIMAX_API_KEY environment variable (shared for both chat and TTS)
• Registers entry in registry.json under category ai
• Updates .gitignore to track minimax/agent-harness/

API Reference

• Chat (OpenAI Compatible): https://platform.minimax.io/docs/api-reference/text-openai-api
• TTS: https://platform.minimax.io/docs/api-reference/speech-t2a-http

Test Results

20 passed in 10.88s (16 unit + 4 e2e, all with real API)

Checklist

• cli-anything-minimax entry point registered
• Default base URL: https://api.minimax.io/v1
• Chat models: MiniMax-M2.7 (default) and MiniMax-M2.7-highspeed
• TTS models: speech-2.8-hd (default) and speech-2.8-turbo
• MINIMAX_API_KEY supported via env, config file, and --api-key flag
• Temperature defaults to 1.0 (MiniMax requirement: must be in (0.0, 1.0])
• TTS audio decoded from hex (SSE stream)
• SKILL.md included for agent discovery
• 20 tests passing (unit + integration)

[yuh-yang]

The PR structure looks good to me, also the e2e test results. But was wondering if there are critical scenarios where such CLIs for calling another LLM API are important 🤔

[omerarslan0]

@octo-patch Went through the diff. Re @yuh-yang's scope question — AnyGen sets precedent for API-wrapping harnesses, and the TTS feature adds real value beyond a simple API client. Fits fine.

Blockers:

1. tts command will crash — The Click parameter output shadows the module-level output() function. When tts() calls output(output_data, ...) on its last line, it tries to call the string "output.mp3" as a function → TypeError. Rename the parameter to output_path.

2. .gitignore missing Step 5 — You have !/minimax/ (Step 4) and !/minimax/agent-harness/ (Step 6) but Step 5 is missing. Add /minimax/* and /minimax/.* in the Step 5 section.

Should fix:

3. handle_error doesn't use functools.wraps — Manually copying __name__/__doc__ works by coincidence of decorator ordering but it's fragile. Use @functools.wraps(func).

4. E2E tests silently fall back to mocks when no API key — This duplicates unit test coverage instead of testing real E2E. Use @pytest.mark.skipif(not API_KEY, ...) instead, so CI doesn't report false passes.

5. registry.json contributor is ximi but PR author is @octo-patch — verify the GitHub handle matches.

Fix 1-2, and this is ready to merge.

[yuh-yang]

Thanks for the MiniMax harness. Two fixes before merge: the REPL should use shlex.split() so quoted chat/TTS prompts work, and models/voices need to respect global --json by returning structured arrays through the shared output helper.

One test cleanup too: test_get_api_key_priority() should mock config loading for the empty-env case so local config files do not affect the test.