ab-scripting-feature-devlisted

Generates an AgenticBrowser Scripting DSL orchestration that implements a feature end-to-end — plan (if needed), TDD implementation, tests, and a lint/style-guide review pass — previews it as docs/scripting-features/feature-<name>.scripting.md, and on user approval runs it via st-eval. Use this whenever the user wants to build or automate a feature "using the scripting DSL", asks to orchestrate coding agents (claude/codex/etc.) to implement something in an AgenticBrowser-based repo, or asks for a multi-agent orchestration script (`AgenticBrowser runBy:` / `scriptBy:` / `groupRunBy:`) for feature development. Trigger on phrases like “Implement it using the scripting DSL,” “Create an orchestration script,” “Create a feature-xxx.scripting.md file”, "scripting DSLで実装して", "オーケストレーションスクリプトを作って", "feature-xxx.scripting.mdを作って", even if the user doesn't say "skill" explicitly.
mumez/pharo-agentic-browser · ★ 3 · Code & Development · score 66

Install: claude install-skill mumez/pharo-agentic-browser

# Scripting Feature Dev Turns a feature request into a runnable AgenticBrowser Scripting DSL orchestration, lets the user preview it before anything executes, and only runs agents against the codebase after explicit approval. ## Reference material This skill depends on the AgenticBrowser Scripting DSL, documented in two files. Use whichever source is freshest: 1. **Bundled snapshot**: `references/scripting-dsl.md` and `references/orchestration-groups.md` next to this SKILL.md 2. **Canonical upstream** (if you have web access and want to double-check for changes): https://github.com/mumez/pharo-agentic-browser/blob/main/docs/scripting.md and https://github.com/mumez/pharo-agentic-browser/blob/main/docs/orchestration-groups.md Read the DSL reference (and the orchestration-groups reference, if the feature looks large enough to need it) before generating a script — the DSL surface changes over time and guessing at syntax produces scripts that fail at runtime instead of at review time. ## Why the preview step matters The generated script spins up real coding agents that write and run code against the repository. Nothing should execute without the user seeing the exact Smalltalk first — that's why every run of this skill produces a markdown file before it produces a side effect. ## Workflow ### 1. Collect feature and goal Ask for (skip whichever the user already gave you): - **Feature** — what should be built or changed - **Goal** — the observable condition that means "d