Bug: Duplicate provider entries in /model picker when providers exist in both built-in and user config

Description

When a provider exists in both:

1. Built-in provider list (HERMES_OVERLAYS, CANONICAL_PROVIDERS, or _PROVIDER_MODELS)
2. User config (providers: section in config.yaml)

The /model picker shows duplicate entries with different names and model counts.

Current Behavior

Example output from Hermes CLI TUI /model command:

Current: deepseek-chat on DeepSeek
› DeepSeek (2 models) ‹ current
OpenRouter (32 models)
Anthropic (8 models)
Z.AI (7 models)
Kimi For Coding (7 models)
OpenAI (0 models) 
openai (0 models)
 deepseek (0 models)
kimi-coding (0 models)

Note the duplicates:

• DeepSeek (built-in) and deepseek (user config)
• Kimi For Coding (canonical) and kimi-coding (user config)
• OpenAI (?) and openai (user config)

Root Cause

The bug is in hermes_cli/model_switch.py::list_authenticated_providers():

1. Section 2 (HERMES_OVERLAYS) adds built-in providers (e.g., slug="deepseek", name="DeepSeek")
2. Section 3 (user-defined endpoints) tries to add user config providers but skips if slug already in seen_slugs
3. Section 4 (custom providers) adds providers from custom_providers list with custom: prefix
   • Checks if slug.lower() in seen_slugs: where slug="custom:deepseek"
   • seen_slugs has "deepseek" but not "custom:deepseek"
   • Result: Adds duplicate custom:deepseek

The deduplication logic fails because:

• Section 4 only checks exact slug match, not base slug (without custom: prefix)
• _pair_key deduplication assumes Section 3 adds all user providers, but Section 3 skips providers already in seen_slugs

Configuration

Relevant config.yaml sections:

providers:
  openai:
    type: openai
    base_url: https://api.openai.com/v1
    api_key_env: OPENAI_API_KEY
    default_flex: True
    models: ['gpt-5.4', 'gpt-5.4-mini', 'gpt-5.4-nano']
  deepseek:
    type: deepseek
    base_url: https://api.deepseek.com
    api_key_env: DEEPSEEK_API_KEY
  kimi-coding:
    type: kimi-coding
    base_url: https://api.moonshot.ai/v1
    api_key_env: KIMI_API_KEY

Expected Behavior

Each provider should appear only once in the /model picker:

• Built-in providers should show with their proper display names
• User config providers that duplicate built-in providers should be suppressed
• Providers should show correct model counts (not 0 for configured providers)

Proposed Fix

In list_authenticated_providers(), Section 4 (lines 1161-1185):

Change from:

if slug.lower() in seen_slugs:
    continue

To:

# Check both the full slug and base slug (without "custom:" prefix)
base_slug = slug[7:] if slug.startswith("custom:") else slug
if base_slug.lower() in seen_slugs:
    continue

Additionally, consider improving the _pair_key logic to handle cases where Section 3 skips providers.

Environment

• Hermes Agent version: (latest main branch)
• Python: 3.11+
• OS: Linux

Steps to Reproduce

1. Add a provider to config.yaml that already exists as a built-in provider (e.g., deepseek, kimi-coding)
2. Run Hermes CLI TUI
3. Type /model
4. Observe duplicate entries

Additional Notes

• The "OpenAI" vs "openai" duplicate might have a different root cause (possibly from Section 1 or label formatting)
• Telegram /model picker works correctly because it uses different code path (_build_model_keyboard in telegram.py)
• This affects user experience by cluttering the model picker with confusing duplicates

Disclaimer: Issue written by Hermes Agent using DeepSeek ⚕️