dsh-taskify: A Task Compiler Plugin for DeepSeek Harness, Turning Rough Ideas into Agent-Ready Tasks
gearvoid/dsh-taskify
A DSH plugin that compiles rough ideas into agent-ready task specs, with one-click refinement and safe backfill.
dsh-taskify is a DSH plugin for DeepSeek Harness that compiles rough requirements into structured task specifications, enabling Coding Agents to execute more accurately.
dsh plugin --profile web add github:gearvoid/dsh-taskify- Category
- Workflow & Automation
- Platform
- DSH Plugin
- Author
- gearvoid
- Distribution
- Plugin
dsh-taskify Key Features
dsh-taskify Repo Summary
dsh-taskify is a Task Compiler plugin (DSH plugin) for DeepSeek Harness, targeting the Web Profile, maintained by GearVoid under the MIT license. It solves the problem of turning a rough, colloquial requirement into a structured task specification that Coding Agents can directly execute. Instead of polishing prompts, it organizes goals, scope, constraints, and acceptance criteria to help agents understand and execute tasks more accurately. Core capabilities include one-click refinement, Literal Lock for protecting literals, undo at any time, limited context understanding, slash command preservation, reuse of the current model, cancellation and retry, and session isolation.
What core features does the dsh-taskify DSH plugin support?
- One-click refinement: Click the "Refine Task" button to the right of the input box to organize the current draft into a structured task specification, including goals, tasks, constraints, and acceptance criteria.
- Literal Lock: Before sending to the model, code blocks, inline code, file paths, URLs, IP/ports, environment variables, version numbers, CLI parameters, and common code identifiers are temporarily locked. After return, they are restored only if the count and order validation passes, preventing the model from altering them.
- Undo anytime: Refined results are only filled back into the input box, never auto-sent; after applying, you can one-click undo to the original draft.
- Limited context: Only uses a small number of recent completed messages from the current session to understand references like "this page" or "the previous change", without reading the workspace or searching the codebase.
- Slash command preservation: Command prefixes like
/planremain unchanged; only the task body after the command is refined. - Reuse current model: Prefers the model selected in the current session, no extra API key configuration required.
- Cancel and retry: You can cancel during refinement, and after a failure, you can retry directly and view error messages.
- Session isolation: Requests, undo points, and error states are isolated per session.
How to install dsh-taskify?
In the DeepSeek Harness Web environment, install the dsh-taskify plugin with the following command:
dsh plugin --profile web add github:gearvoid/dsh-taskify
After installation, start or restart DeepSeek Harness Web:
dsh web
The "Refine Task" button will appear to the right of the input box, before the model selector and send button.
How to use dsh-taskify?
- Write a task draft in the input box.
- Click "✨ Refine Task".
- Review the auto-filled task specification.
- Send manually, or click "↶ Undo" to restore the original text.
Button states include: disabled when input is empty, refinable, refining (cancellable), applied (undoable), edited after apply (can refine again), and failed (can retry).
What about dsh-taskify's security and privacy design?
- Does not read workspace files or search the codebase.
- Does not read
.env, SSH keys, or local credential files. - Does not access additional third-party services.
- Does not auto-send or submit refined tasks.
- Recent conversation is kept as limited text and filters common credential patterns.
- Drafts containing Reference Chips are not refined in the current version.
Which DeepSeek Harness versions and platforms does dsh-taskify support?
Verified on DeepSeek Harness 0.1.0-rc.6, supporting the Web Profile. It does not modify the DeepSeek Harness core or Agent Presets. Since DeepSeek Harness is still in rapid iteration, future versions may require adaptation.
