**WORKFLOW SKILL** — Produce a reusable SKILL.md from a conversation, PR, or code change. Outputs a conservative draft, concise clarifying questions, and runnable example prompts tailored to this repository.
Installation
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
**WORKFLOW SKILL** — Produce a reusable SKILL.md from a conversation, PR, or code change. Outputs a conservative draft, concise clarifying questions, and runnable example prompts tailored to this repository.
user-invocable
true
metadata
{"scope":"user","version":"0.2.0"}
create-skills
Short summary
Generate a safe, discoverable SKILL.md from a conversation, issue, or PR. The agent extracts the step-by-step workflow, decision points, inputs/outputs, and success criteria; surfaces ambiguous items as clarifying questions; and emits 3–5 runnable example prompts. Drafts are conservative and require explicit confirmation to write into the workspace unless askConfirm=false is provided.
When to use
When a conversation, PR, or code change describes a repeatable multi-step workflow that should be reusable.
When you want a clear checklist for contributors (e.g., adding scenes, wiring audio, updating dialogue trees) rather than ad-hoc instructions.
Inputs
conversation (optional): text or highlighted excerpt to extract the workflow from. If omitted, agent scans recent chat context.
scope (optional): user or workspace (default: user). Stored under metadata.scope in drafts.
skillName (optional): short, hyphenated name for the skill (default: create-skills).
targetPath (optional): explicit path to write the SKILL.md instead of the default.
askConfirm (optional, default true): if true, present the draft and request confirmation before writing.
draftOnly / toPrompt (optional): return a draft only, or generate a .prompt.md instead of a full SKILL.
Outputs
A conservative SKILL.md draft (shown inline). If confirmed, writes to the chosen path:
workspace: .agents/skills/<skillName>/SKILL.md
user prompt: VS Code prompts folder as <skillName>.prompt.md when toPrompt=true.
A short list of clarifying questions for ambiguous or missing parts.
3–5 runnable example prompts tailored to the repository.
Process (detailed)
Gather context: use the provided conversation or read recent chat context and referenced files (e.g., dialog_uebersicht.md, src/store.ts, audio.ts).
Extract the workflow: enumerate ordered steps, actors, side effects, required files, inputs/outputs, and decision points.
Identify quality checks and completion criteria that define "done" for the workflow.
Draft SKILL.md using the template below and place any ambiguity in Clarifying questions.
Show the draft if askConfirm=true or draftOnly=true.
If the user confirms, write the file via apply_patch; otherwise iterate.
Offer next steps: create a feature branch, commit, and open a PR with the new skill.
Decision points & branching
Multi-step workflows → full SKILL.md with steps, branching logic, and example prompts.
Single-step templates → recommend a .prompt.md and offer to create it.
Team-wide changes or CI hooks → prefer workspace scope and recommend adding documentation to AGENTS.md.
Quality criteria / completion checks
Frontmatter: name and description present; user-invocable: true when intended for direct invocation.
metadata for non-standard fields like scope and version.
description includes trigger phrases (e.g., "Use when:") for discoverability.
Process lists steps, decision points, and explicit success criteria.
Examples: 3–5 runnable example prompts with consistent parameter naming (camelCase by default).
Safety: draft is not written automatically unless confirmation disabled.
YAML: no tabs, quote values that include colons.
Frontmatter example
---name:create-skillsdescription:'Use when: convert a conversation or PR into a reusable SKILL.md'user-invocable:truemetadata:scope:userversion:'0.2.0'tags:-skills-workflow---
SKILL.md template (canonical)
---
name: {{skillName}}
description: 'Use when: concise trigger phrases that enable discovery'
user-invocable: true
metadata:
scope: {{user|workspace}}
version: '0.1.0'
---
# Short summary
# When to use
# Inputs
# Outputs
# Process
# Decision points
# Example prompts
# Clarifying questions
Example prompts (Neurotoxic-tailored)
Scaffold a new scene (includes the 5 required wiring steps):
Returns: reproduction steps, likely root causes (e.g., store persistence, combineItems ordering), minimal code-edit suggestions or diff snippets, and test ideas.
Create a single-step prompt (no multi-step workflow):
