Foreword

The ecosystem philosophy of DeepSeek Harness (DSH) emphasizes modularization, and skills are its core assets. When maintaining a skill library, developers often face inconsistent documentation structure, missing quality standards, and irregular naming. The dsh-skill-authoring plugin aims to address these pain points by providing standardized scaffolds, quality-gate auditing, and automatic validation to ensure the skill library is clean and usable.

Plugin Positioning

This is a DSH plugin maintained by boomzikazita for creating and maintaining high-quality agent skills. It does not rely on complex configuration and directly extends DSH’s toolset through the plugin mechanism.

Core Features

The plugin provides the following main tools and mechanisms:

  1. skill_scaffold
    Generates a standardized SKILL.md scaffold for a new skill, including frontmatter, whenToUse, Boundaries, QuickRef, and Examples placeholders, and automatically creates references/index.md.

  2. skill_audit
    Audits an existing SKILL.md file and outputs pass/fail status against the 9-point quality gate standard.

  3. skill_check_name
    Validates whether the skill name conforms to DSH conventions (regular expression ^[a-z0-9]+(-[a-z0-9]+)*$), allowing only lowercase letters, digits, and hyphens.

  4. skill-write-guard
    Installs a write hook. When using the write or edit command to modify a SKILL.md file under ~/.dsh/skills/, the system automatically runs an audit and issues a warning if it does not pass the quality gate.

9-Point Quality Gate Standard

The plugin adapts the vibe-coding-cn standard for DSH and includes the following nine checks:

  1. name: conforms to the regular expression and has length ≤ 64 characters
  2. description: contains “what + when”, length 20–1024 characters
  3. whenToUse: contains ≥ 3 executable triggering conditions
  4. boundaries: contains ≥ 3 “Not For / Boundaries” statements
  5. quickref: contains 1–20 directly reusable patterns (quick reference)
  6. examples: contains ≥ 3 end-to-end examples (including Input / Steps / Acceptance)
  7. references: references/index.md file exists
  8. verification: uncertain claims must be marked with source/evidence level
  9. style: uses an operations-manual style (short imperative sentences, not documentation dumping)

Installation and Enablement

The plugin is installed via the DSH command line. After installation, restart the DSH instance to register the tools.

dsh plugin --profile web add github:boomzikazita/dsh-skill-authoring

After installation, restart the corresponding dsh web instance or profile to ensure the new tools take effect.

Configuration and Dependencies

The plugin requires no additional configuration (config: {} in cordis.patch.yml is sufficient). The following describes the runtime environment:

  • Skills directory: defaults to ~/.dsh/skills; the user’s home directory is dynamically obtained at runtime, with no hardcoded path.
  • Python environment: the plugin depends on a Python 3.8+ interpreter to run the quality gate engine (scripts/skill_gate.py).

Typical Usage

  • Generate scaffold: use the skill_scaffold command to initialize a new skill.
  • Quality check: use skill_audit to check whether an existing skill document meets the 9-point standard.
  • Format validation: use skill_check_name to confirm the skill name format is correct.
  • Automatic protection: when editing skill files, skill-write-guard automatically intervenes with an audit.

Ecosystem and License

The plugin follows the MIT License. The community directory skillhub.cn provides an index page for this plugin, but the plugin itself has no official affiliation with DeepSeek / High-Flyer.