AI Agent Hub
Back to plugins
🧠

dsh-skill-indexer

Memory Updated 2026.08.20

Run the following command in DeepSeek Harness:

dsh plugin install NattoCB/dsh-skill-indexer

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install NattoCB/dsh-skill-indexer in the DeepSeek Harness terminal to install this plugin; the source repository is available at https://github.com/NattoCB/dsh-skill-indexer .

About this plugin

When local SKILL.md files grow to dozens or even hundreds, an agent must rely on manual lookup or hardcoded paths every time it needs a specific skill. dsh-skill-indexer turns that into plain language: say something like "pull this week ETF data and chart it", and the plugin routes to the etf skill through a two-level (category then skill) recall index, with clear hit, low-confidence, and miss fallback states. You no longer need to remember every skill name.

Under the hood it scans SKILL.md files under ~/.workbuddy/skills and $DSH_HOME/skills, parses frontmatter triggers and keywords, groups them into categories, then ranks within each category using keyword coverage weighted by a pos_weight formula. The entire scan, parse, index, and route pipeline relies only on Node built-in modules (fs, path, crypto) with zero third-party runtime packages. Source directories are strictly read-only; the index updates incrementally by content hash, preserving usage counters for unchanged skills. Integration surfaces include the skill_index model tool, the /skill-index CLI command, and the cordis.patch.yml config section, plus optional categories.yaml overrides for manual category assignment.

Built for DSH users who have accumulated a large local skill library and want agents to reach the right skill through a one-sentence intent. Fully local, read-only and safe, zero external dependencies, and works out of the box without maintaining any additional search service.

Use Cases

  • An agent must locate the right skill among dozens of local SKILL.md files without manual enumeration
  • Describe a task in one natural-language sentence and get the best-matching skill back with hit / low / miss confidence
  • A growing skill library needs a single read-only recall entry point that replaces hardcoded paths and external search services

Best For

  • DSH users with a large local SKILL.md collection who want agents to auto-match skills
  • Developers who prefer fully local, read-only, zero third-party runtime dependencies
  • Ops teams in security-sensitive environments that require skill search running without any external network calls