DeepSeek Harness (DSH) developers often need to reproduce sessions, validate results, or extract reusable skills from successful cases. Manually processing exported files is not only cumbersome, but also makes it difficult to guarantee the completeness and traceability of evidence. DSH Session Lab addresses this need by providing standardized workflow tools.

What It Is

DSH Session Lab is a standalone, pre-1.0 community DSH bundle. It is not an official DeepSeek product; it is maintained by developer zhangguiping-xydt. The plugin registers three complementary workflow skills in the DSH runtime directory for handling session evidence, skill extraction, and trajectory comparison.

Core Features

The plugin provides the following three core capabilities:

  1. Evidence Packaging and Validation ($dsh-capsule)
    Packages DSH session exports into portable evidence bundles (.dshc) and validates their integrity. The generated bundle includes a manifest, sanitized session events, integrity hashes, and optional workspace patches.

  2. Skill Extraction and Evaluation ($dsh-teach)
    Extracts reusable skills from successful sessions and performs independent baseline/treatment evaluations to validate the effectiveness of the generated skills.

  3. Controlled Trajectory Comparison ($dsh-time-machine)
    Compares the trajectories of two sessions, starting from their common completed turn, to understand the causal boundaries of where differences arise.

Installation and Enablement

Installing the plugin requires the DSH CLI. Make sure DeepSeek Harness is installed and the corresponding profile (for example, web) is configured.

dsh plugin --profile web add github:zhangguiping-xydt/dsh-session-lab

To ensure reproducible installation, you can lock a specific version tag:

dsh plugin --profile web add github:zhangguiping-xydt/dsh-session-lab#v0.1.0

After installation, restart the DSH profile and check the configuration to ensure the plugin is correctly mounted:

dsh --profile web --dump-config | rg '# == dsh-session-lab'

Usage Examples

After installation and profile restart, you can use the following commands by invoking the Skill directly in the conversation:

  • Package evidence:
    Use $dsh-capsule to package and verify this DSH export.
  • Extract skills:
    Use $dsh-teach to extract and independently evaluate a Skill from this successful session.
  • Compare trajectories:
    Use $dsh-time-machine to compare these two sessions from their common completed turn.

Before running, be sure to read the relevant SKILL.md documentation to understand input requirements, output content, safety checks, and known limitations.

Environment Requirements

Running this plugin depends on specific environment versions:

  • Node.js: 22.19+ or 24+, used for DSH and plugin installation.
  • Python: 3.10+, used for skill helper scripts.

Notes

  • Unofficial product: DSH Session Lab is an independent community project and is not affiliated with DeepSeek or High-Flyer.
  • Security: Raw export files, patches, images, and reports usually contain sensitive content. Automatic replacement modes are not an anonymization mechanism. SHA-256 integrity checks are only used to verify that data has not been tampered with; they cannot verify the publisher’s identity.
  • Functional limitations: The plugin does not add a Host session import endpoint, does not execute capsule content, and does not claim to perform exact model replay.

Summary

DSH Session Lab provides DeepSeek Harness users with a standardized workflow for extracting value (evidence, skills) from sessions and analyzing differences (trajectory comparison). For workflow developers who need to reproduce, share, or optimize agent behavior, it is a practical toolset. For more details, see the GitHub repository or the community directory.