Preface

The plugin ecosystem of DeepSeek Harness (DSH) emphasizes breaking down models, capabilities, and settings into composable plugins; the community directory page is an independent site and should not be equated with the DeepSeek / Hypersphere official app store.

If you need to integrate ChatGPT Codex in DSH and wish to manage login, model catalog, Fast models, 1M / context picker rows, as well as optional Search and Image capabilities in a single plugin, NOirBRight/dsh-llm-codex provides these capabilities.

The following introduces its positioning, core capabilities, installation methods, and typical usage.

What is this

dsh-llm-codex is a DeepSeek Harness plugin maintained by NOirBRight, positioned as:

ChatGPT Codex login, sortable catalog, and Fast models for DeepSeek Harness

It contributes the Codex card under Settings → LLM Providers, and the Host-side Codex llm route.

It is important to note that the LLM Providers page and shared order store are solely owned by dsh-llm-providers-ui; this plugin only contributes the Codex card and Host llm route.

Core Capabilities

Verified capabilities of dsh-llm-codex include:

  1. ChatGPT Codex login and OAuth session management.
  2. The Codex card in the LLM Providers settings page.
  3. The card displays usage limits.
  4. Model catalog stored in settings.models.
  5. Default catalog includes:
    • gpt-5.6-sol / gpt-5.6-sol-fast
    • gpt-5.6-terra / gpt-5.6-terra-fast
    • gpt-5.6-luna / gpt-5.6-luna-fast
  6. Fast rows send service_tier: priority.
  7. Supports 1M rows and -<n>k / -<n>m context picker rows.
  8. Default reasoning effort can be edited per model row.
  9. Optional Search, view_image, and codex_generate_image capabilities, disabled by default.
  10. Registers optional Search / Image adapters when dsh-model-switch v0.2+ is present.

Additionally, verified resources explicitly state:

  • There is no apiKeyEnv.
  • There is no user-editable base URL.
  • It does not read or write ~/.codex/auth.json.
  • It does not share the credential file with dsh-codex-connect.

Installation and Enabling

First, confirm the DSH version. The README requires DeepSeek Harness 0.1.0-rc.6 or higher; the Release installation section states the artifact target is DeepSeek Harness 0.1.2-alpha.1.

Install a specific version from GitHub:

dsh plugin --profile web add github:NOirBRight/dsh-llm-codex#v0.3.7

If you want to use the Settings → LLM Providers page in the Web interface, you need to install dsh-llm-providers-ui simultaneously. Headless Host routes may not require an Owner.

Login and Sessions

Go to Settings → LLM Providers → Codex, click Sign in with ChatGPT to start the official ChatGPT OAuth flow.

Sessions are only stored on the Host:

$DSH_HOME/codex-oauth.json

The file permissions are 0600. Clicking Sign out will delete this file.

After login is complete, the Codex card will display usage limits.

Model Directory

The model directory uses settings.models. The default catalog contains six rows:

gpt-5.6-sol / gpt-5.6-sol-fast
gpt-5.6-terra / gpt-5.6-terra-fast
gpt-5.6-luna / gpt-5.6-luna-fast

Fast rows send:

"service_tier": "priority"

The 1M row is not in the default catalog but can be added from the official picker, for example:

gpt-5.6-sol-1m

The Picker id can also use context suffixes like -<n>k or -<n>m, for example:

gpt-5.6-sol-272k
gpt-5.6-sol-1m

The plugin strips the suffix and sets the DSH compaction budget.

The default reasoning effort can be edited per model row.

Optional Capabilities

Search, view_image, and codex_generate_image are disabled by default.

After enabling any capability and clicking Save, it will register or unregister immediately without a restart.

Verified Search behaviors include:

  • Search model dropdown defaults to gpt-5.6-luna.
  • Search modes are cached, indexed, and live.
  • Default mode is cached.
  • Search does not write to web.searchProvider.
  • Search does not write to agent-default-model.

codex_generate_image uses Codex usage, typically about 3–5 times that of a single text turn. Generated files are located by default at:

generated-images/

Unless a path is set.

Model Switch Integration

When dsh-model-switch v0.2+ is present, Codex registers optional Search / Image adapters.

Verified resources explicitly state:

  • Model Switch integration does not register a Vision adapter.

Errors and Session States

Verified failure states include:

  • Failure when Chat has no session: MISSING_CREDENTIAL.
  • Failure when refreshing a stored session: AUTH.

Applicable Scenarios and Notes

Suitable use cases:

  • Integrating ChatGPT Codex in DSH.
  • Needing to maintain Codex login, usage limits, and model catalog uniformly.
  • Needing Fast rows, 1M rows, or -<n>k / -<n>m picker rows.
  • Needing optional Search, view_image, or codex_generate_image capabilities.

Inappropriate or non-dependent scenarios:

  • Do not want to directly read ~/.codex/auth.json.
  • Do not want to share the credential file with dsh-codex-connect.
  • Expect the plugin to provide a user-editable base URL or apiKeyEnv.

Before installing, check the source code and license. This plugin’s license is MIT. The plugin runs with the current dsh process permissions; it is recommended to confirm the source, version, and behavior match expectations.

When using the Web interface, the LLM Providers page and shared order store are still owned by dsh-llm-providers-ui; this plugin only contributes the Codex card and Host llm route.

Conclusion

dsh-llm-codex provides ChatGPT Codex login, model catalog, Fast model rows, 1M / context picker rows, and optional Search and Image capabilities (disabled by default) for DeepSeek Harness.

GitHub:

https://github.com/NOirBRight/dsh-llm-codex

Directory Page:

https://www.skillhub.cn/plugins/NOirBRight/dsh-llm-codex