skill-creator-advanced — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited skill-creator-advanced (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
此 skill 的目標是把「做 skill」變成可重複執行的工程流程,而不是一次性的 prompt 雜談。
它同時提供:
1) 先從現有對話、repo、範例任務整理 2-3 個 use cases,不夠再補問。 2) 為每個 use case 寫 trigger 語句與 done looks like。 3) 建立 skill 資料夾:
python scripts/init_skill_advanced.py <skill-name> --path <output-dir>4) 補完新 skill 的 SKILL.md,優先寫對 YAML 的 description。 5) 做格式與結構檢查:
python scripts/format_check.py <path/to/skill>
python scripts/quick_validate.py <path/to/skill>6) 規劃真實測試案例,必要時產生測試計畫:
python scripts/generate_test_plan.py <path/to/skill> --out references/test_plan.md7) 準備 eval workspace,讓 with-skill / baseline 能沿用固定目錄結構:
python scripts/prepare_eval_workspace.py <path/to/skill>8) 打包成 .skill:
python scripts/package_skill.py <path/to/skill> <output-dir>9) 若要優化 description 的觸發品質,另外準備 trigger eval set,再跑:
python scripts/run_eval.py --eval-set <path/to/trigger-evals.json> --skill-path <path/to/skill> --model <model-id>
python scripts/run_loop.py --eval-set <path/to/trigger-evals.json> --skill-path <path/to/skill> --model <model-id> --apply-best當使用者要建立或改版 skill 時,請用下列順序推進;可以跳步,但要明確說明原因。
1) Phase 0:從上下文萃取需求
2) Phase 1:需求與 use cases
3) Phase 2:架構與 SKILL.md
scripts/、references/、assets/。description 必須同時回答兩件事:這個 skill 做什麼、什麼情況下應該觸發。description 要用真實使用者語句,而不是作者自嗨式分類。SKILL.md,細節與變體移到 references/。4) Phase 3:撰寫指令
5) Phase 4:格式檢查與最小合規驗證
format_check.py 修掉結構與格式問題。quick_validate.py 做最小合規確認。6) Phase 5:測試、evals 與 benchmark
zh、en、mixed、縮寫/俗稱。assets/evals/evals.json。<skill-name>-workspace/iteration-N/,每個 eval 各自有 with_skill/ 與 baseline 目錄。scripts/aggregate_benchmark.py 彙整 benchmark,再用 scripts/generate_review.py 產生 review viewer。scripts/check_regression_gates.py 檢查是否達到發版門檻。7) Phase 6:打包與發布
package_skill.py 產生 .skill。8) Phase 7:迭代與維護
完整細節見:
references/lifecycle.mdreferences/testing-playbook.mdreferences/description-optimization.mdreferences/eval-workflow.mdreferences/eval-schemas.mdreferences/multilingual-trigger-strategy.mdreferences/skill-boundary-management.mdreferences/regression-gates.mdreferences/skill-roi-model.mdreferences/distribution-playbook.mdreferences/patterns-troubleshooting.md1) 先把 description 寫對
2) 先從上下文學會,再提最少的問題
3) 把脆弱步驟移到 scripts
4) 避免 context 膨脹
SKILL.md 放流程與導航。references/,必要時再讀。5) 測試要真實,不要只測漂亮案例
6) with-skill 與 baseline 要用同一批 evals 比
7) 先處理 skill 邊界,再處理 wording
8) ROI 不成立的 skill 不值得硬留
9) 不要在 skill folder 放 README.md
scripts/init_skill_advanced.py:建立帶測試/發布欄位的 SKILL.md 骨架。scripts/format_check.py:格式與結構檢查器(含 --fix)。scripts/quick_validate.py:最小合規驗證。scripts/generate_test_plan.py:產生測試計畫模板。scripts/prepare_eval_workspace.py:從 assets/evals/evals.json 建立 iteration workspace。scripts/aggregate_benchmark.py:彙整 with-skill / baseline run 結果,輸出 benchmark.json 與 benchmark.md。scripts/check_regression_gates.py:依 benchmark 與門檻設定判斷是否可發版。scripts/run_eval.py:跑 description trigger eval,輸出 query-level 與 run-level 診斷結果。scripts/improve_description.py:依 trigger eval 失敗型態重寫 description,保留 transcript。scripts/run_loop.py:把 eval 與 description 改寫串成多輪迭代,可選擇直接套用最佳 description。scripts/generate_report.py:產生 description optimization 的 HTML 報告。scripts/utils.py:共用的 SKILL.md / JSON 讀寫輔助。scripts/package_skill.py:驗證後打包成 .skill。assets/evals/evals.json:保存真實測試 prompt、預期輸出與 expectations。assets/evals/regression_gates.json:保存 benchmark 的發版門檻設定。scripts/generate_review.py:把 workspace 結果輸出成 review HTML。<skill-name>-workspace/iteration-N/:保存每輪 with-skill / baseline 的輸出、grading 與 benchmark。交付給使用者時,通常包含:
SKILL.md + scripts/ + references/ + assets/)assets/evals/evals.json<skill-name>-workspace/iteration-N/ 的 benchmark 與 review 輸出.skill 打包檔~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.