boomzikazita/dsh-skill-lazy
DSH skill catalog lazy loading: name+one-line summary injection, on-demand search (~500 tokens)
dsh-skill-lazy saves tokens by replacing DSH's full skill-catalog injection (name + up to 500-character description per skill, roughly 3K tokens per session with dozens of skills) with a compact catalog of just names and one-line summaries (≤ 40 characters by default), sorted by call frequency. Full descriptions stay one skill_search call away: the tool queries the full catalog and returns hits grouped by MECE domains with full descriptions, usage counts and a domain-coverage report that flags empty or weak domains. A tools/post-execute hook records every skill tool invocation into usage.json to drive ordering, and an optional proactive hint surfaces top-3 relevant skills.
Install
dsh plugin --profile web add github:boomzikazita/dsh-skill-lazyGitHub install per README (EN primary): dsh plugin --profile web add github:boomzikazita/dsh-skill-lazy. npm dsh-skill-lazy 404 verified 2026-09-03 — GitHub route only.
Compatibility
DSH profiles with the skill-catalog message and skills seam; runtime usage.json written by the post-execute hook.
Details
- Repo: boomzikazita/dsh-skill-lazy
- Category: Agent Capabilities
- Stars: 0
- Version: git github:boomzikazita/dsh-skill-lazy (npm 404 verified)
- Last push: 2026-08-18
- First seen: 2026-08-18
Recent updates
No changelog section in README.
FAQ
- How much context does it save?
- Roughly 3K tokens of persistent context per session with dozens of skills installed, by injecting only names plus one-line summaries instead of full descriptions.
- Can the agent still find skills?
- Yes — a skill_search tool queries the full catalog on demand and returns hits grouped by MECE domains with full descriptions and usage counts.
- How is the compact order chosen?
- Entries are sorted by call frequency (most-used first) tracked via a post-execute hook into usage.json, falling back to alphabetical order for ties.
Alternatives
ben7am1n/dsh-review-skills · hexbee/dsh-skill-panel