القائمة
Author a new agent-cloud skill — scaffold a skill package (a directory with a valid SKILL.md and any helper scripts). Use when the user asks to create, scaffold, write, or design a new skill.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
| name | skill-creator |
| description | Author a new agent-cloud skill — scaffold a skill package (a directory with a valid SKILL.md and any helper scripts). Use when the user asks to create, scaffold, write, or design a new skill. |
| version | 1.0.0 |
Use this when the user wants to create a new skill for this agent platform. Your job is to
produce a skill package: a directory named after the skill, containing a SKILL.md (and any
helper scripts the skill needs), written into the working directory with the file tools
(write_file / edit).
A skill is a folder with a SKILL.md. Once a user installs and enables it for an agent, the
platform copies it into the agent's working directory at .skills/<name>/ and injects only the
skill's name + description + location into the system prompt. The agent then reads the
SKILL.md on demand (with read_file) when a task matches the description — so the body can be
as long as needed without costing context until it's actually used (progressive disclosure).
Implication: the description is the only routing signal the model sees up front. Everything
else lives in the body and is paid for only when read.
A SKILL.md is YAML frontmatter (between --- lines) followed by a Markdown body.
Frontmatter fields:
name (required) — lowercase, matches ^[a-z0-9][a-z0-9._-]*$, no .., ≤ 64 chars. Must equal
the directory name. Examples: notion, pdf-tools, gh-issues.description (required, non-empty) — one or two sentences: what the skill does AND when to use
it, with concrete trigger words. This is what makes the model pick the skill, so be specific.
Bad: "Helps with files." Good: "Convert and inspect PDFs (extract text, split, merge). Use when
the user works with .pdf files."version (optional, string, default "0.0.0") — e.g. "1.0.0".requires (optional mapping) — declared dependencies, e.g. requires: { bins: [pandoc] }. Use it
to document the CLIs/packages the skill expects in the sandbox.Body guidance (the on-demand instructions):
bash tool over prose. Show real examples
and expected output.Put any scripts/templates the skill needs inside the skill directory alongside SKILL.md. After
install they live at .skills/<name>/... in the working directory, so reference them by that path,
e.g. bash .skills/<name>/scripts/run.sh. Make scripts self-contained (no absolute paths; install
deps with pip install --user / npm install -g as the environment requires).
name (kebab-case, matches the regex above).SKILL.md:
write_file to "<name>/SKILL.md" with the frontmatter + body."<name>/scripts/..." if needed.--- lines.name matches ^[a-z0-9][a-z0-9._-]*$, ≤ 64 chars, and equals the directory name.description is present, non-empty, and says what + when.<available_skills> on the next turn. (Alternatively: upload the folder as a .zip
if archive upload is enabled.)pdf-tools/SKILL.md:
---
name: pdf-tools
description: "Extract text from PDFs and split/merge pages. Use when the user works with .pdf files."
version: "1.0.0"
requires:
bins: [python3]
---
# pdf-tools
Extract text:
pip install --user pypdf
python3 -c "import pypdf,sys; print('\n'.join(p.extract_text() for p in pypdf.PdfReader(sys.argv[1]).pages))" input.pdf
Split/merge: see pypdf's `PdfWriter`. Always confirm the output file exists before reporting done.
name, valid pattern, ≤ 64 chars.description says what it does and when to use it, with trigger words..skills/<name>/....