鮮度
Agent skills break when the tools, SDKs, or models they target change. Each skill in this catalog has a freshness score calculated from when its substantive instructions or scripts were last updated in Git and whether people are still using it.
Tiers
New skills without a proven user base show 🌱 New instead of a percentage. Once a skill has real usage, it receives a score from 0% to 100%. Retired skills move to the 墓場.
| Badge | Score | Stage | Meaning |
|---|---|---|---|
| 🌱 New | Unproven | Incubating | Recently added skill that does not have enough user activity yet to earn a percentage score. |
| ✨ Mint | 90 – 100% | Gold standard | Recently updated in Git with a stable or growing user base. |
| 🌿 Fresh | 70 – 89% | Active | Either recently updated with a smaller user base, or an older skill with steady weekly usage. |
| 🕸️ Stale | 50 – 69% | Aging | No recent commits while weekly usage is slipping below its historical baseline. |
| 🤢 Spoiled | 25 – 49% | Decaying | Outdated last commit paired with falling weekly usage. |
| 🦠 Rotten | 0 – 24% | Abandoned | Untouched for months with zero weekly activity after initial adoption. |
| 🪦 R.I.P. | Retired | Graveyard | Removed from the catalog and listed in the 墓場. |
Formula
Each proven skill combines substantive Git commit age (U, 45-day half-life) with usage (V) across kungfu CLI telemetry, npx skills installs, and web interactions:
U = 0.5 ^ (age_days / 45)
V = log_norm(kungfu_cli + npx_skills + web_activity)
Vitality = round(80 × (1 - (1 - U) × (1 - 0.85 × V × U^0.25)))
Adoption = round(20 × V × U^0.35)
Score = clamp(0, 100, Vitality + Adoption - Penalty)
- No-op commit filter: Repo-wide metadata edits, license header sweeps, and directory moves are ignored when computing
age_days. - Data sources: Combines
kungfuCLI telemetry (loads,runs,learnssynced to Firestore), rawnpx skillsinstalls fromskills.sh, and web engagement.
Security
Each skill page displays results from three security scanners run by skills.sh: Gen Agent Trust Hub, Socket, and Snyk.
- Prompt safety: Gen Agent Trust Hub and Socket check for prompt injection, credential leaks, and unsafe dependencies.
- Static scanner notes: Snyk flags markdown files that document running local CLI tools like
go testoruv. Each flagged skill includes a note explaining why the alert triggered.