AI Agent Hub
Back to plugins
⚙️

dsh-exec-extension

Workflow Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install LvDAO/dsh-exec-extension

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

Install by running dsh plugin install LvDAO/dsh-exec-extension in DeepSeek Harness; the source repository is at https://github.com/LvDAO/dsh-exec-extension .

About this plugin

When you need to run a DeepSeek task inside a CI pipeline or an automation script, the stock headless mode lacks fine-grained control over sandboxing, approval policies, and output formatting, and model overrides depend on on-disk configuration that is hard to isolate per-process. dsh-exec-extension solves this by shipping a dedicated exec profile that executes a one-shot, fully headless task in a single command, from structured input to programmatic output.

The CLI follows the interaction model of OpenCode run and Pi -p: piped stdin is merged into the prompt, @file attaches file text, --format json streams session events as JSONL, and --dir sets the working directory. Sandbox and approval are granular, with three permission levels (read-only, workspace-write, danger-full-access) paired with three approval policies (ask, never, allow), so a CI job can enable --full-auto for a zero-interaction run. Model overlay runs in-process via Rust WASM and never touches disk configuration, making it natural for concurrent or containerized deployments.

It is built for engineers wiring LLM tasks into CI/CD, automation teams that batch-invoke DeepSeek through scripts, and agent-workflow developers who want a complete CLI control surface without modifying headless-runner. By design it does not ship MCP, HTTP serving, or session resume, keeping it lightweight, single-shot, and predictable.

Use Cases

  • Running a one-shot LLM task in a CI pipeline with structured output
  • Batch-invoking DeepSeek from scripts with sandbox and approval control
  • Overlaying model config in-process inside containers without disk writes

Best For

  • Engineers wiring LLM tasks into CI/CD pipelines
  • Automation teams that batch-invoke DeepSeek through scripts
  • Agent-workflow developers wanting full CLI control without modifying headless-runner