| Native Hermes Agent pain | After installing the family |
|---|---|
| You re-teach workflows every session; context swells with redundant instructions. | Hub auto-routes to the right skill in seconds; mechanisms load on demand. |
| No security baseline — SQL injection, hardcoded secrets slip through. | Three-layer defense: silent disciplines + commit gate + on-demand deep scan. |
| Cross-session context loss — "we did this before" forgotten. | Dual-layer memory: project snapshots + experience/vulnerability libraries. |
| Quality depends on mood — no review, no gate. | AI code review engine + dual-gate before commit. |
| Token waste — huge skill bodies always loaded. | Progressive disclosure: ≤5-line contracts in context, full body on demand. |
Paste this one line to your agent:
Or manually:
mkdir -p ~/.hermes/skills/vibe-coding-family
git clone --depth 1 https://github.com/yuchen1017918/vibe-skill-ops /tmp/vso
cp -r /tmp/vso/vibe-coding-family ~/.hermes/skills/
L1 organization form → L2 dev mode → L3 scenario grid → L4 executing skill. Hub loads on trigger words; policies inject silently; memories persist across sessions.
30 routing skills (tool×15 / workflow×7 / policy×6 / meta×2) + 6 hubs. 8 deprecated archived (files kept, no routing).
Three-layer defense + 4 non-negotiable red lines + vulnerability memory (CWE-classified). Real SQLi interception tested.
snapshot-notes v2.0 (project state + experience extraction) + dev-memory + vuln-memory. Cross-session continuity.
Official-format adaptation: 36 skills, description ≤500 chars, auto-generated by convert-dsh.py. Install to ~/.agents/skills/.
README / CONTRIBUTING / SECURITY all bilingual (EN default + zh-CN), LANG switch links at top.
workflow-distillation turns your described workflow into a skill; user layer overrides standard layer.
GitHub · README (EN) · README (中文) · Tutorial · Contributing · Security · DeepSeek Harness