Preface¶

The DSH plugin ecosystem emphasizes “everything is a plugin”; the community directory is an independent site with no official affiliation to DeepSeek / High-Flyer, and should not be understood as an official app store. For agent developers, the problem that learning-type plugins need to solve is not “generating another batch of notes,” but turning a real input into traceable evidence, updating learning state, preserving rollback boundaries, and allowing the primary agent to continue handling the current response.

gitlearnos provides the official DeepSeek Harness Developer Preview native bundle, exclusive to GitLearnOS. It stores the learner-owned learning memory in a private local Git repository, and provides bounded read-only observation, a queue panel, and authorized write-back transactions within the Harness.

What This Is¶

gitlearnos is maintained by Guojiz and licensed under MIT. It is positioned as Evidence-driven learning memory and automation for Git-native AI agents.

It is designed for Git-native agents: the primary agent can identify useful learning events, connect evidence to learning goals, provide next-step actions, and leave checkable, rollback-capable records. Learning can still happen in classrooms, from books, on practice platforms, in projects, or through interactions with other AIs; gitlearnos does not move all of learning into a single application—it only connects evidence useful for the learner’s next decisions.

Core Capabilities¶

The following verified capabilities are described below.

• Provides the official DeepSeek Harness Developer Preview native bundle, exclusive to GitLearnOS.
• Stores learner-owned learning memory in a private local Git repository.
• Offers bounded read-only observation of learning_status and learning_route.
• Supports learning_apply transactions authorized by gitlearnos.yml, applying typed events, knowledge-gap, model, review, and dashboard operations in a single rollback-capable Git commit.
• Provides a loopback-only, read-only panel displaying the agent-maintained Next up queue, respecting Panel: expand|collapse.
• Supports five queue actions, placing review, practice, close-with-one-question, ask-a-teacher, or read-notes requests into the session input.
• Supports safe-auto, preview, and manual write-back modes.
• Supports an optional local RAG layer, with RAG-Anything explicitly supported as the first option.

Write-Back and Permission Boundaries¶

learning_apply is not an ordinary write. It requires checks for learner identity, setup/config, base revision, and write-authority; learning_record is retained as a compatibility wrapper.

Write-back modes can be distinguished as follows:

1. safe-auto: Executes minimal safe write-backs when the goal, evidence, learning objectives, and privacy boundaries are clear.
2. preview: Proposes precise changes but does not write.
3. manual: Waits for approval.

Additionally, maintenance and due-review are only complete after tasks have been created and observed in the real scheduler.

Typical Usage¶

Start with the core-ready Quickstart¶

Use the single canonical setup prompt from the core-ready Quickstart. It will ask the agent to:

1. Identify private goals.
2. Confirm learning objectives and subjects against current materials.
3. Recommend local RAG.
4. Wait before deployment.
5. Detect actual capabilities and report the undo boundary.

Advance with Real Learning Events¶

A single question, an answer, a page photo, class notes, teacher feedback, practice results, or recurring difficulties can all serve as learning events.

When executing according to learning goals and real input, the flow can be written as:

goal and real input
→ organize traceable evidence
→ ask from the current gap
→ keep the answer or external feedback
→ recheck independently later
→ update state with one reversible Git commit

The corresponding steps are: organize traceable evidence, ask from the current gap, keep the answer or external feedback, recheck independently later, and update state with one rollback-capable Git commit.

When Larger Materials Are Needed¶

For larger school textbooks, course packs, notes, or persistent personal knowledge, a local RAG layer can be recommended, with RAG-Anything as the first explicitly supported option. RAG provider access is an optional independent layer and is not built into this bundle.

Visual Input and Multimodality¶

The DeepSeek default provider is text-only. If image processing is required, a third-party multimodal model can be configured, or an authorized vision/OCR bridge plugin can be installed. If neither is available, the agent will request a transcript rather than guess at unseen content.

Installation and Enablement¶

Verified materials do not provide official installation commands. To integrate, it is recommended to rely on the bundle and documentation in the repository, and not to piece together installation commands based on the plugin name alone.

Applicable Scenarios and Notes¶

gitlearnos is suitable for DSH developers or learning agent scenarios that require:

• Storing learning state in a private local Git repository by default.
• Writes as single rollback-capable Git commits with explicit permission checks.
• A loopback-only, read-only panel to view the Next up queue.
• An optional local RAG layer for textbooks, course packs, notes, or persistent personal knowledge.

Notes before use:

• The plugin runs with the current dsh process permissions; review the source code and license before installation.
• A private repository is recommended by default; remotes such as GitHub are optional.
• The panel is loopback-only, read-only, and labels development sample data.
• Verified materials do not provide star counts or official installation commands; do not supplement uncertain data based on this.

Links¶

• GitHub: https://github.com/Guojiz/gitlearnos
• Plugin directory page (plugin lead): https://www.skillhub.cn/plugins/Guojiz/gitlearnos