dsh-skill-recommender: DeepSeek Harness Skill Recommender
The `dsh-skill-recommender` plugin in DeepSeek Harness analyzes local session records (such as DSH, Codex, and Claude) to build a user profile, then combines local and remote skill directories and uses a weighted model to score and rank skills, recommending the best-matching open-source skills to users. It requires no API key and runs entirely locally, while allowing users to adjust weights and matching indices for filtering through a Web interface. It also provides command-line tools for scanning, generating recommendations, and checking status, with strict permissions on configuration files.
Read Moredsh-skillradar: Recommending the Next Skill to Load for the Current DSH Session
dsh-skillradar is an MIT-licensed plugin for DeepSeek Harness designed to solve the problem of models struggling to determine which skill to load in multi-skill sessions. The plugin scans the skills visible in the current session, scores them based on recent conversation text (English and Chinese token overlap), and recommends the next most relevant skill. It provides a `skill_radar` model tool with support for `session_id` and `limit` configuration, and optionally offers an interactive Web UI. In terms of data security, it only reads the most recent 8000 characters of message text in the current session and the skill directory, performing no data writes, network requests, or credential usage. Installation requires compatibility with DSH 0.1.0-rc.6; the plugin can be added via GitHub or a local clone, and a restart is required for it to take effect.
Read More