Skill Creator

Skill Creator (AgenticX)

Guide for turning a successful conversation workflow into a discoverable skill under ~/.agenticx/skills/.

For listing, installing, or registry operations, use the separate agenticx-skill-manager skill.

When to use

• User says: 「落盘 skill」「封装成 skill」「工具调用太多，保存成 skill」
• You completed a non-trivial task (roughly 5+ tool calls) and the steps should be reused
• An existing agent-created skill needs a clearer description or body after skill_manage patch

Capture intent (from current chat)

Before writing anything, extract from this conversation:

1. Goal — what the skill enables the agent to do
2. Trigger — phrases/contexts when the skill should activate (put in description, not only the body)
3. Inputs/outputs — formats, files, APIs, edge cases the user corrected
4. Proven steps — the sequence that actually worked (not aspirational steps)

Confirm the skill name (hyphen-case, e.g. a-stock-daily-report) with the user if unclear.

SKILL.md anatomy

my-skill/
├── SKILL.md          # required
├── scripts/          # optional — deterministic helpers
└── references/       # optional — long docs loaded on demand

Frontmatter (required)

---
name: my-skill          # MUST match skill_manage name= parameter
description: One line on what it does AND when to trigger it (pushy but accurate).
---

• name — lowercase hyphen-case identifier; required for Desktop Skills scan
• description — primary trigger signal; include when-to-use phrases
• Optional: version, title, requires_tools, metadata — AgenticX allows these; do not omit name

Keep the body under ~500 lines; split heavy detail into references/.

Persist (mandatory path)

1. skill_use(skill-creator) — you are here; follow this guide
2. Draft full SKILL.md text (frontmatter + body)
3. skill_manage(action='create', name='<dir-name>', content='<full SKILL.md>')
   • Or from_path / from_url for large files
   • Never bash_exec / file_write directly into ~/.agenticx/skills/
4. Read the tool result:
   • discoverable: true → skill appears in Settings → Skills
   • frontmatter_fixed non-empty → tell the user what was auto-corrected (e.g. injected name)
   • ERROR → do not claim success; fix and retry
5. skill_list — optional double-check the name appears

Writing quality

• Imperative steps; concrete commands and paths
• Include failure modes you hit in this session (proxy issues, missing deps, etc.)
• No conversation transcript dumps — actionable procedure only
• Security: no exfiltration, no misleading or destructive instructions

Minimal template

---
name: example-skill
description: Do X when the user mentions Y or Z. Use for ...
---

# Example Skill

## When to use
- ...

## Steps
1. ...
2. ...

## Verification
- How to confirm success

Patch existing skills

skill_manage(action='patch', name='...', old_string='...', new_string='...')

After patch, verify discoverable: true in the response before telling the user it is visible in Settings.