AI Agent Hub
Back to plugins
⚙️

dsh-github-companion

Workflow Updated 2026.09.11

Run the following command in DeepSeek Harness:

dsh plugin install vclike/dsh-github-companion

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install vclike/dsh-github-companion in DeepSeek Harness to install; the source is hosted at https://github.com/vclike/dsh-github-companion

About this plugin

AI agents that need to talk to GitHub often end up wrapping the gh CLI, which drags in PATH requirements, version drift, and sandbox incompatibility. dsh-github-companion sidesteps all of that by speaking REST and GraphQL natively over fetch, delivering a complete, zero-dependency GitHub integration that behaves identically on Windows, Linux, and macOS.

The plugin mounts as three cordis bundles: 33 registered tools grouped into read and discovery (always on), issue writes, git-data writes, repo creation, and local clone, each switchable with conservative defaults; an in-process permission gate that lets you pick from three modes and three actions to precisely scope write access; and an Actions cost guard that refuses pushes or releases while in-progress workflow runs exist, plus a per-tag cooldown to stop debugging loops from burning billed minutes. Multi-file pushes are a single atomic git-data round-trip, repo creation is always private, and the token never touches argv, the remote URL, or logs.

It is built for DeepSeek Harness users who want their agent to handle code summarisation, issue management, PR work, release cutting, and fork syncing inside the conversation, while fine-grained PAT permissions and the permission gate keep write risk tightly scoped.

Use Cases

  • Summarise repo changes, manage issues, and cut releases entirely inside the chat
  • Scope AI write access precisely with a three-mode, three-action permission gate
  • Refuse pushes when in-progress workflow runs exist and enforce a per-tag cooldown to stop billing loops

Best For

  • DeepSeek Harness users who want their agent to operate GitHub in-conversation
  • Engineering teams that need fine-grained PAT plus permission-gate control over agent writes
  • Developers who want cross-platform consistency and sandbox safety without a gh CLI dependency