skill-creator
Create, revise, evaluate, or package Agent Skills. Use for SKILL.md instructions, activation descriptions, and bundled resources.
v2026.11.13-21cd5683 התקנות+3 ב-7 ימים39 כוכביםMITסונכרן לפני 11 שעות
התקנה
עדיין אין לכם את ה-CLI של NotHoly? התקינו אותו קודם.
$ curl -fsSL https://notholy.com/install.sh | sh$ notholy skills install aeondave/skill-creatorנכתב אל ~/.notholy/skills אחרי בדיקת החתימה וה-SHA-256.
חתימהחתום, ed25519
חבילה0e554123…a4fbbcc7
0e55412348d2…7b63a4fbbcc75 קבצים · 7.9 KBסריקהנמוךממצא אחד
- קישורים לדומיינים חיצוניים
מקורgithub.com/AeonDave/malskill/tree/main/knowledge/skill-creator@21cd568
הפריט הזה עדיין לא נכתב בעברית, ולכן הוא מוצג בשפה אחרת: English.
בעמוד הזה
Produce a skill that changes how an agent handles a recurring task. Keep only instructions, resources, and constraints that contribute to that outcome.
Establish the gap
Use the request, existing skill, and applicable repository instructions to identify the missing behavior and what completion requires. Inspect only resources relevant to the change. Ask for missing information only when it changes the scope, contract, or authorization.
For a skill-app with executable workflows and consistent deliverables, load skill-app-creator for application contracts, runtime diagnostics, and recovery design.
- Reuse an existing skill when its scope already fits; fix its routing if discovery is the problem.
- Create a skill only for a distinct recurring capability. Keep one-off inputs in the task prompt and repository-wide conventions in repository instructions.
- Preserve supported behavior, metadata, user decisions, and authorization boundaries unless their change is part of the request.
- Separate the observed failure from a proposed remedy. Try deleting or narrowing conflicting instructions before adding another rule.
Author the smallest useful instruction set
Brief, clear, specific, useful. Assume competence in ordinary domain knowledge and tool use. Retain non-obvious constraints, decision criteria, output contracts, and verification that prevent a concrete failure. Ground technical claims in the current implementation, a real run, or a primary source; qualify uncertain conditions.
Define the outcome and when work is complete. Prescribe a sequence only when order matters; otherwise let the agent choose the method. Record genuine approval boundaries without inserting review stops into already-authorized work. Do not turn a workaround for one model into a universal requirement.
Frontmatter
---
name: my-skill
description: "Capability and the specific task that should activate it."
---name: matches the folder, lowercase letters/digits/hyphens, at most 64 characters; no leading, trailing, or consecutive hyphens.description: the shortest clear routing signal, at most 1024 characters. Lead with the distinctive task. Include a near-miss boundary only when needed; omit workflow summaries, broad topic associations, synonym lists, and activation pressure.- Add optional metadata only when useful. Declare non-obvious runtime requirements in
compatibility; keep product-specific behavior scoped to that runtime.
Load references/spec.md when choosing optional fields or checking specification details. The field limits are ceilings, not writing targets.
Body and resources
Keep shared decisions and constraints in SKILL.md. For independently used workflows, make it a compact router with explicit conditions for loading each resource. A short, single-workflow skill can remain one file.
| Resource | Add when |
|---|---|
scripts/ |
Repeated mechanics or fragile operations need deterministic execution. Specify inputs, outputs, dependencies, and failures. |
references/ |
A subtask needs depth that other tasks can skip. Link the exact file with a "load when" condition. |
assets/ |
The agent needs a reusable template or static file in the output. |
Keep each fact in one canonical location. Delete irrelevant material instead of moving it into a reference. Omit generic tutorials, design defenses, historical notes, and extra README/CHANGELOG/install guides. Add examples only to resolve ambiguity; add navigation when a long reference cannot be scanned easily.
Load references/patterns.md when deciding how to split workflows or repair resource routing.
Create or revise
For an existing skill, edit in place; do not re-scaffold. In malskill, repository maintenance tools live in the repository's scripts/ directory and are not bundled with this skill. For a new skill in that repository:
python scripts/init_skill.py <skill-name> --path <output-dir>
# Request resource directories only when needed:
python scripts/init_skill.py <skill-name> --path <output-dir> --resources referencesUse --examples only when sample files help; replace or remove them before finishing. Execute new or changed helpers on representative inputs, including relevant failure cases. In malskill, executable repository-tooling changes need focused tests under tests/; standalone skills and other repositories use their available host or repository tooling.
Evaluate the change
Use the smallest check that can detect the claimed improvement:
- Editorial change: review the diff and validate structure.
- Instruction or workflow change: try a realistic task in a clean context and inspect the result and actions taken.
- Substantial behavior change: compare the candidate with the pre-edit or no-skill baseline using the same inputs and environment. Set expected outputs and prohibited outcomes before grading.
- Routing change: exercise natural requests and plausible near misses with the available skill catalog.
Load references/pressure-testing-skills.md when designing behavior comparisons or diagnosing failures. Load references/skill-triggering-tests.md when tuning activation.
Test the models and hosts used by the intended audience when making compatibility claims; name untested targets. Check for unnecessary reads, tool calls, tests, and approval stops as well as incorrect outputs. A shorter file is not evidence of better behavior.
Validate and finish
In malskill, scripts/quick_validate.py requires PyYAML for strict frontmatter parsing; install it in the runtime before running the validator. Standalone skill packages do not include repository maintenance tools.
In malskill, run these commands from the repository root for each changed skill directory. In other repositories, use the available validators and check the same structural conditions:
python scripts/quick_validate.py <skill-dir>
python scripts/sweep_skills.py <skill-dir>
python scripts/check_changed_files.pyResolve frontmatter errors, broken links, scaffold markers, and workstation-path leakage. The sweep enforces broken links through its exit status; inspect its report-only findings too. These checks establish structure and hygiene, not behavioral quality.
Finish the requested edits, affected checks, and corrections before handing back the result. Report what was tested and what remains unverified. In malskill, package with python scripts/package_skill.py <skill-dir> only when an archive is requested; use python scripts/validate_all.py <repo-root> only for repository-wide validation. Run these commands from the malskill repository root.
After a real failure or model/runtime change, reassess the affected instructions and rerun relevant scenarios. Keep a rule only if it still improves the outcome.
הקבצים בגרסה הזו
references/412.4 KB
- patterns.md2.5 KB
- pressure-testing-skills.md4.0 KB
- skill-triggering-tests.md2.4 KB
- spec.md3.5 KB
- SKILL.md7.1 KB
הרשאות
| הרשאה | הוצהר | נמצא בסריקה |
|---|---|---|
| קורא קבצים | לא | לא |
| כותב קבצים | כן | כן |
| מריץ פקודות shell | כן | כן |
| משתמש ברשת | כן | כן |
| משתמש במחברים | לא | לא |
| שולט בדפדפן | לא | לא |
ממצאי הסריקה
נבדק ב-13 בנוב׳ 2026
- מידעקישור · קישורים לדומיינים חיצוניים
סריקה נקייה מורידה את הסיכון, אבל לא מוכיחה שהחבילה בטוחה. כדאי לקרוא מה מתקינים.
| גרסה | פורסמה | Commit | SHA-256 | חתימה | שינוי בהרשאות |
|---|---|---|---|---|---|
| v2026.11.13-21cd568 | 13 בנוב׳ 2026 | 21cd568 | 0e554123 | חתומה | – |
עוד מ-AeonDave
הכול מהמפרסם הזהעוד בקטגוריה דוחות ותאימות
לקטגוריה- skillduty1g/skill5
- rea-ci-tests-auditmorluto/rea-ci-tests-audit4
- tracking-threat-actor-infrastructuremukul975/tracking-threat-actor-infrastructure4
- performing-authenticated-scan-with-openvasmukul975/performing-authenticated-scan-with-openvas4
- slicing-code-contexttrailofbits/slicing-code-context4
- 1337aeondave/13374
שאלו כל מה שהסקופ מאפשר.
יוצרים חשבון, מוסיפים קרדיטים ושואלים את השאלה הראשונה בצ׳אט או בטרמינל.
- 1/3
יצירת חשבון
פרטי או חברה. אפשר לעבור לחשבון חברה אחר כך בלי לאבד יתרה או מפתחות.
פרטיחברה - 2/3
הוספת קרדיטים
בכרטיס אשראי. תשלום לפי טוקן, בלי מנוי.
$5.00טעינה מינימלית - 3/3
שאלה ראשונה
בצ׳אט, או להתקין את ה-CLI ולהריץ notholy login בטרמינל.
$ notholy login