skill-creator

name	skill-creator
description	Create or update InternAgent skills. Use this skill when the user wants InternAgent to create a reusable skill, turn a repeated workflow into a SKILL.md package, improve an existing skill, or decide what scripts, references, assets, and triggering description a skill should include.

Skill Creator

Use this skill to create, improve, or review InternAgent skills.

InternAgent skills are reusable instruction packages. A skill is a directory that contains a required SKILL.md file and optional supporting resources:

skill-name/
+-- SKILL.md
+-- scripts/
+-- references/
`-- assets/

Creation Workflow

1. Understand the user's intent before writing files.

   • What recurring task should the skill help with?
   • When should InternAgent use this skill?
   • What inputs, outputs, tools, credentials, or external services are involved?
   • Are there examples, edge cases, or success criteria?

2. Choose the skill shape.

   • Use only SKILL.md for instruction-only workflows.
   • Add scripts/ when repeated code, deterministic parsing, file conversion, or API calls would otherwise be rewritten each time.
   • Add references/ for detailed documentation, schemas, policies, API notes, or long examples that should be loaded only when needed.
   • Add assets/ for templates, sample files, images, or other files used to produce outputs.

3. Pick a stable skill name.

   • Use lowercase letters, digits, and hyphens only.
   • Keep the name short and specific.
   • Make the directory name match the name frontmatter value.

4. Write SKILL.md.

   • Include YAML frontmatter with name and description.
   • Put trigger guidance in description, because InternAgent uses it before loading the body.
   • Keep the body focused on procedures, constraints, examples, and how to use bundled resources.
   • Prefer clear, general guidance over one-off details copied from the current conversation.

5. Validate the result.

   • Confirm SKILL.md exists and frontmatter parses.
   • Confirm required fields are non-empty.
   • Confirm referenced files actually exist.
   • Run or syntax-check scripts when the skill includes them.
   • Test with at least one realistic user prompt when practical.

Writing Guidance

Write skill descriptions that include both what the skill does and when it should trigger. A good description is specific enough to avoid accidental use, but explicit enough to trigger when the user phrases the request differently.

Good pattern:

---
name: report-builder
description: Create structured research or experiment reports from notes, data files, logs, or previous chat context. Use when the user asks InternAgent to produce a reusable report, summarize experiment outcomes, compare runs, or turn raw project material into a polished document.
---

Avoid putting "when to use this skill" only in the body. If the trigger condition is not in description, InternAgent may not load the skill.

Keep SKILL.md concise. Move long API documentation, schemas, policies, or examples into references/ and link them from the main file with clear instructions about when to read them.

Safety And Maintainability

• Do not create skills that hide behavior from the user, exfiltrate data, capture credentials, or bypass authorization.
• Do not store secrets in a skill. Describe the required environment variable or configuration instead.
• Do not add unnecessary README, changelog, or installation guide files unless the user explicitly asks for them.
• Prefer updating an existing relevant skill over creating near-duplicates.
• If a skill is generated from a one-time workflow, generalize it so it remains useful outside the current thread.

InternAgent Placement

For project-provided skills, place the skill as a first-level directory under skills/, for example:

skills/my-skill/SKILL.md

For user-imported or generated runtime skills, prefer .internagents/imported-skills/my-skill/ so generated content does not accidentally become repository source.

When a skill should be active by default, add its relative path to skills.selected in deepagent.config.json.