dsh-task-planner DSH plugin

dsh-task-planner adds a condition-reflex planning system to DeepSeek Harness. It maintains an experience library of past tasks as Markdown lessons, recalls them via token matching, and uses LLM evaluation to decide whether to reuse or replan. Every plan auto-drafts a lesson, and the agent updates it with outcomes at task close.

skills-workflows

How to install the dsh-task-planner DSH plugin

dsh plugin --profile web add github:<your-user>/dsh-task-planner

This source command needs manual review. Copying does not run it.

dsh-task-planner DSH plugin data source

dsh-task-planner DSH plugin snapshot date: Aug 16, 2026

discovered

What the dsh-task-planner DSH plugin can do

Experience library with save, recall, and list commands, storing lessons as plain Markdown with signature keywords.
Condition-reflex planning via `plan_task` that recalls past lessons, evaluates fit with LLM, and produces decomposed steps mapped to capabilities.
LLM-driven decision-making: the model uses past experiences and an optional capability catalog to decide reuse or fresh planning.
De-AI deliverable standard: any textual output step must include a humanize-then-review pass before delivery.
Auto-persist: `plan_task` drafts a lesson automatically, and the agent marks it verified with the outcome at loop close.
Zero keys and zero absolute paths: all paths are configurable, default experience library at `~/.dsh/planner-lessons`.

Where the dsh-task-planner DSH plugin fits

Plan complex multi-step tasks by recalling similar past solutions and adapting them.
Build a reusable knowledge base of task plans that improve over time through feedback.
Automate the process of documenting lessons learned from each task execution.
Enforce a human-review step before delivering any generated textual output (docs, scripts, etc.).
Integrate with a capability catalog to match each step against available skills and plugins.

Who the dsh-task-planner DSH plugin is for

Developers using DeepSeek Harness who want to manage complex task workflows.
AI agents (or users) needing a structured, experience-driven planning approach.

dsh-task-planner DSH plugin limitations

Requires the `llm`, `shell`, and `tools` services, which are standard in DSH but must be present.
The model call uses the harness default model; reasoning models may need a generous `maxTokens` (8k used internally).
Experience library is stored as local Markdown files only; no remote or shared storage is provided.
The plugin is in early development (no version mentioned) and may have limited testing.

dsh-task-planner DSH plugin: from the repository README

Quoted from the ztl34245881-commits/dsh-task-planner README, the upstream source of the dsh-task-planner DSH plugin. Copyright remains with the original authors.

Task planning with **experience muscle-memory** for [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) (`dsh`). Give a task → the agent recalls **past similar solutions** (condition reflex), evaluates whether they fit, and produces a dynamic plan matched against its capabilities — **never hard-coded combos**. Every plan auto-drafts a lesson into the experience library; when the task closes, the agent updates the outcome. The more you work, the smarter the reflex. ## Features - 🧠 **Experience library** (`task_memory save/recall/list`): persistent lessons as plain Markdown with signature keywords. Recall uses a 2–3-char sliding-window tokenizer, so "weekly report" still hits a "daily report" lesson. - ⚡ **Condition-reflex planning** (`plan_task`): recall → LLM evaluates fit (reuse & improve, or explain why not and plan fresh) → decomposed steps with capability matching → risks → next actions. - 🤖 **LLM-driven, not rule-driven**: the model decides what to use per task; the plugin only supplies context (past experiences + optional capability catalog). - ✍️ **De-AI deliverable standard**: any textual output step (docs/sheets/slides/copy/scripts) must include a huma

Read the full READMERepository license: MIT

dsh-task-planner DSH plugin questions

How do I install dsh-task-planner?

Run `dsh plugin --profile web add github:<your-user>/dsh-task-planner` in your terminal. Replace `<your-user>` with the actual GitHub username or organization. Alternatively, you can clone the repo and add it as a local bundle using `dsh plugin --profile web add /path/to/dsh-task-planner`.

Can I change the location of the experience library?

Yes, you can configure the `lessonsDir` in your profile's `cordis.patch.yml` under the plugin's config section. By default it is `~/.dsh/planner-lessons`. For example: `config: { lessonsDir: /custom/path/lessons }`.

What happens if the LLM cannot find a matching past lesson?

The plugin will still plan fresh. The `plan_task` command recalls past lessons, but if none match or the LLM evaluates that none fit, it will generate a new plan from scratch. The new plan will be drafted as a lesson automatically, so the experience library grows over time.

How do I see all stored lessons?

Use the `task_memory list` command in the DSH environment. It will show all lessons stored in the experience library. You can also use `task_memory recall { task }` to search for specific lessons by keywords.

Does this plugin require an API key or external service?

No, the plugin is zero-key. It uses the LLM service already configured in your DSH profile (the default model). You only need to ensure the `llm`, `shell`, and `tools` services are available. No additional API keys are needed for the plugin itself.