skill-author
Guides an agent through writing a good SKILL.md, covering the name, a description that triggers at the right time, concise instructions, bundled scripts and tests. Use when the user asks to create, improve or review an agent skill, or wants to turn a repeated task into a reusable
#agents #prompt-engineering
Install
npx skills add https://github.com/Lethe044/agentskill-kit/tree/main/skills/skill-author
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install lethe044-agentskill-kit@llmmart
git clone https://github.com/Lethe044/agentskill-kit.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole lethe044/agentskill-kit collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Skill author
A skill is a folder with a SKILL.md file. The agent reads the name and description of every installed skill
and loads the full instructions only when a task matches. The description is therefore the most important line.
When to use
- The user keeps repeating the same kind of task and wants it captured.
- The user has a draft skill that never triggers, or triggers at the wrong time.
- A skill needs a review before it is shared.
Workflow
- Collect real examples. Ask for two or three actual requests the skill should handle, and one it should not.
- Pick the name. Lowercase letters, digits and single hyphens, at most 64 characters, same as the folder name.
- Write the description. Third person, what it does and when to use it, under 1024 characters, no angle brackets. Include the words a user would actually type.
- Write the instructions. Short sections: when to use, steps, output format, one example, errors.
Keep the file under 500 lines and move long reference material into
references/. - Add a script only when code is more reliable than prose (parsing, counting, validating). Use the standard library, read arguments or standard input, print JSON, and exit with a non-zero code on failure.
- Add tests as a
cases.jsonfile inside atestsfolder: one happy path, one edge case, one error case at minimum. - Check it. Run
askit lint,askit scanandaskit test, thenaskit score.
Description checklist
- Does it start with a verb that says what the skill does?
- Does it contain "Use when ..." with concrete triggers?
- Would a different skill be a better match for any of these words?
- Is it written as "Analyzes logs" and not "I analyze logs"?
Writing instructions
- Prefer steps an agent can verify over general advice.
- Tell the agent what to do when input is missing or unclear.
- Show the output format once, with a real example.
- Do not repeat what the model already knows. Spend words on what is specific to this task.
- Never ask the agent to hide actions from the user or to ignore other instructions.
Layout
my-skill/
SKILL.md
scripts/ optional code the skill runs
references/ optional long documents, linked from SKILL.md
tests/ cases.json or test_*.py
evals/ evals.json for checking model behaviour
Errors
If the user cannot give examples, write the description from the stated goal and mark the skill as a draft in
metadata.
Files (agentskill-kit)
-
SKILL.md 2.9 KB
--- name: skill-author description: Guides an agent through writing a good SKILL.md, covering the name, a description that triggers at the right time, concise instructions, bundled scripts and tests. Use when the user asks to create, improve or review an agent skill, or wants to turn a repeated task into a reusable skill. license: MIT metadata: version: 1.0.0 author: Lethe044 tags: [skills, authoring, prompt-engineering, agents] --- # Skill author A skill is a folder with a `SKILL.md` file. The agent reads the name and description of every installed skill and loads the full instructions only when a task matches. The description is therefore the most important line. ## When to use - The user keeps repeating the same kind of task and wants it captured. - The user has a draft skill that never triggers, or triggers at the wrong time. - A skill needs a review before it is shared. ## Workflow 1. **Collect real examples.** Ask for two or three actual requests the skill should handle, and one it should not. 2. **Pick the name.** Lowercase letters, digits and single hyphens, at most 64 characters, same as the folder name. 3. **Write the description.** Third person, what it does and when to use it, under 1024 characters, no angle brackets. Include the words a user would actually type. 4. **Write the instructions.** Short sections: when to use, steps, output format, one example, errors. Keep the file under 500 lines and move long reference material into `references/`. 5. **Add a script only when code is more reliable than prose** (parsing, counting, validating). Use the standard library, read arguments or standard input, print JSON, and exit with a non-zero code on failure. 6. **Add tests** as a `cases.json` file inside a `tests` folder: one happy path, one edge case, one error case at minimum. 7. **Check it.** Run `askit lint`, `askit scan` and `askit test`, then `askit score`. ## Description checklist - Does it start with a verb that says what the skill does? - Does it contain "Use when ..." with concrete triggers? - Would a different skill be a better match for any of these words? - Is it written as "Analyzes logs" and not "I analyze logs"? ## Writing instructions - Prefer steps an agent can verify over general advice. - Tell the agent what to do when input is missing or unclear. - Show the output format once, with a real example. - Do not repeat what the model already knows. Spend words on what is specific to this task. - Never ask the agent to hide actions from the user or to ignore other instructions. ## Layout ``` my-skill/ SKILL.md scripts/ optional code the skill runs references/ optional long documents, linked from SKILL.md tests/ cases.json or test_*.py evals/ evals.json for checking model behaviour ``` ## Errors If the user cannot give examples, write the description from the stated goal and mark the skill as a draft in `metadata`.
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.