dsh-py-codeact
Run the following command in DeepSeek Harness:
dsh plugin install CNSeniorious000/dsh-py-codeact
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install CNSeniorious000/dsh-py-codeact in your DeepSeek Harness terminal; source code at https://github.com/CNSeniorious000/dsh-py-codeact
About this plugin
DSH's built-in Code Mode spawns a fresh stateless worker per call, so variables, loops, and intermediate results cannot survive between turns. This plugin collapses the action space to a single tool: python, backed by a persistent IPython session where state carries across calls, while every other harness tool is bridged as an awaitable under dsh.tools for the model to import on demand. This is the CodeAct shape in its purest form: one tool, one session, everything else a callable function.
IPython magics, top-level await, transform_cell, execution history, and the native traceback formatter all come for free. Each bridged tool is a real async def whose keyword-only signature and return type (named TypedDicts, unions, Literals) are rendered from the tool's own schema by dsh's SDK renderers, so what the model reads in the prompt and what it gets from introspection are the same text from the same source. MCP transport envelopes are unwrapped before they reach the cell, and each server can be imported as a module. Exclusive mode (the default) filters the model-facing list down to python alone, with a guard that names the correct import route when the model tries to call anything else directly.
Built for Python-first agent developers who need cross-call state; experimental or research workflows that lean on IPython magics, top-level await, or long-running sessions; and teams mounting multiple MCP servers that want them to feel like native Python modules inside the reasoning loop. The longer and more stateful the session, the more the plugin pays for itself.
Use Cases
- Multi-turn Python agents that need variables, loops, and intermediate state to survive across turns
- Experimental workflows that lean on IPython magics, top-level await, or transform_cell
- Mounting multiple MCP servers and treating each one as a native Python module with per-tool imports
Best For
- Python-first agent developers who need cross-call state persistence
- Research teams building long sessions that depend on IPython advanced features
- Workflow builders who want MCP tools presented in prompts with async def signatures and TypedDict return types
Related Plugins
A method pack that makes AI coding agents plan against your real baseline, prove completion with fresh evidence, and reduce reworks and unsafe changes.
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.