Skip to content

v1.0.0

Latest
Latest

Choose a tag to compare

View all tags
@tiefeiyu tiefeiyu released this 14 Aug 09:22
· 1 commit to main since this release
v1.0.0
59caac8

dsh-see-image v1.0.0

First public release of dsh-see-image — a see_image tool plugin for DeepSeek Harness that gives text-only sessions the ability to "see" images through any OpenAI-compatible vision model.

What it does

  • see_image(file_path, question?) — reads a PNG/JPEG/WebP/GIF through DSH's file sandbox and returns a text description
  • Copilot zero-config mode — point baseURL at https://api.individual.githubcopilot.com and it uses gpt-4.1 vision with automatic token refresh (works on the Copilot Free plan)
  • Generic provider mode — any OpenAI-compatible VLM (OpenAI, Ollama, vLLM, LM Studio, …) via baseURL + model + apiKeyEnv, or keyless for local servers
  • Configurable maxTokens, timeoutMs, maxBytes; sandbox-native file handling with fs observation
  • Single-file plugin (~250 lines), MIT license

Install

See the README. Mount index.js in ~/.dsh/profiles/web/cordis.patch.yml:

plugins:
  - id: dsh-see-image
    name: /absolute/path/to/dsh-see-image/index.js
    config:
      baseURL: https://api.individual.githubcopilot.com
      model: gpt-4.1
      apiKeyEnv: VISION_API_KEY

New cordis.patch.yml entries hot-reload without a host restart.

What's included

  • Plugin source (index.js)
  • Copilot login + token-refresh helper scripts (scripts/)
  • Mount smoke test (test/mount-test.mjs)
  • CI workflow (.github/workflows/ci.yml)

Assets

  • dsh-see-image-1.0.0.zip — full repository snapshot
Assets 3
Loading