Add NanoGPT API support (Issue #129)

[robbee184-spec]

Summary

Implement native NanoGPT API provider support so users can call NanoGPT directly without routing through OpenRouter.

API contract

• Canonical base URL: https://nano-gpt.com/api/v1
• Chat endpoint: POST /chat/completions
• Model discovery: GET /models
• Authentication: Authorization: Bearer <key>
• Environment variable: NANOGPT_API_KEY
• Wire format: OpenAI-compatible
• Static fail-open model: minimax/minimax-m2.7 (live discovery remains authoritative)

Changes

• Adds nanogpt to the LLM provider/configuration surface and fallback chain.
• Routes chat through a NanoGPT-specific OpenAI-compatible adapter with direct Bearer authentication.
• Enables live NanoGPT model discovery through the canonical models endpoint.
• Adds generated-env, shell setup, interactive setup, and Universal API Config coverage.
• Redacts NanoGPT keys from exported configuration and server-visible text.
• Strips NANOGPT_API_KEY before spawning local Codex/Claude/Hermes processes.
• Preserves the newer Hugging Face provider added to main after the original PR head.

Verification

Exact head f60f1abf93bdfd159609c8488c4cf041db45b468:

• npm run typecheck — pass
• npm test — pass (65 files, 682 tests; repository self-checks pass)
• npm run lint — pass with 0 errors (133 existing warnings)
• npm run build --silent — pass
• git diff --check — pass

The fallback model and its context/output/capability metadata were checked against the live GET /api/v1/models?detailed=true response. The implementation does not use or route through OpenRouter.

Fixes #129

[jmagly]

This does not currently implement NanoGPT’s published API contract, and the exact head is not build-green.

• npm run typecheck fails at src/__tests__/nanogpt-provider.test.ts:54 because cfg is declared but unused.
• NanoGPT documents the OpenAI-compatible base URL as https://nano-gpt.com/api/v1, not https://api.nanogpt.com/verse1: https://docs.nano-gpt.com/
• The documented environment variable is NANOGPT_API_KEY, not RBE_API_KEY: https://docs.nano-gpt.com/authentication
• Model IDs should come from GET /api/v1/models; the PR hard-codes rbe-1 without support from the published model contract: https://docs.nano-gpt.com/api-reference/endpoint/models
• Please also add nanogpt to the export-redaction invariant and env template/setup path; the current redaction test omits the new provider, so future config refactors could expose or silently mishandle this key.

Please correct the endpoint/key/model contract and make the tests validate the published interface rather than the current self-consistent constants.

[jmagly]

Maintainer courtesy update at f60f1abf93bdfd159609c8488c4cf041db45b468:

I rebased the provider work onto current main and corrected the reviewed NanoGPT contract:

• canonical direct base URL: https://nano-gpt.com/api/v1;
• canonical key name: NANOGPT_API_KEY;
• direct OpenAI-compatible POST /chat/completions with Bearer authentication;
• live OpenAI-compatible GET /models discovery;
• removed the unsupported rbe-1 assumption and replaced it with the currently documented/live minimax/minimax-m2.7 fail-open entry, with metadata verified from models?detailed=true;
• added generated-env, interactive setup, and Universal API Config coverage;
• added export redaction, sk-nano-* response redaction, and local-agent child-environment isolation for the NanoGPT key;
• preserved the newer Hugging Face provider changes that caused the old head to conflict.

Exact-head local verification:

• npm run typecheck — pass
• npm test — pass (65 files, 682 tests; all repository self-checks pass)
• npm run lint — pass with 0 errors (133 existing warnings)
• npm run build --silent — pass
• git diff --check — pass

I force-updated only after confirming the contributor branch still matched the reviewed old head eaedc21d46528693f79f2b3af4ca7759dbb6aed9, using an exact lease. Please review the replacement commit; no unrelated branch content was removed beyond rebuilding the obsolete single commit on current main.

Superseded by corrected head f60f1ab. The requested NanoGPT endpoint, key, model-discovery, redaction, setup, and exact-head verification corrections are complete; repository CI run 30597775219 passed.