Turn an idea or objective into a goal package for /goal. Interviews the user, builds a reviewed fact sheet via Plannotator, then explores the codebase to produce an execution plan.
A structured workflow that turns an idea or objective into a goal package under goals/<slug>/ for use with the /goal command. It combines an interactive user interview, a reviewed fact sheet built through the Plannotator browser session, and codebase exploration to produce an execution plan. Working JSON files in the goal directory hold provenance and iteration state, while the markdown files become the human-readable authoritative goal package.
The process runs in phases. First, Rearticulate restates what the user wants in two or three sentences, doing minimal shallow codebase exploration if the goal is vague, and waits for confirmation before creating the goal directory. An optional one-at-a-time grilling pass (adapted from the MIT-licensed grill-me skill) can resolve interdependent decisions before building a compact interview bundle. The Interview Bundle phase packages only high-leverage questions, each with a recommended answer and options, into interview.json; questions answerable by exploring the codebase are resolved by exploration instead of being asked. The Fact Sheet phase derives simple, testable, verifiable facts from the interview results into facts-review.json, each flagged for whether automated verification is recommended and preselected.
Interview and facts steps are launched as monitored foreground Plannotator commands (plannotator setup-goal interview and plannotator setup-goal facts, with --json) whose results are written verbatim to interview-result.json and facts-result.json. A strict browser-session patience rule governs the workflow: the agent must wait for the user to submit, dismiss, or stop, and must never close, kill, restart, refresh, or open a second session just because the UI appears idle; revisions are made by editing the working JSON and launching a fresh command. The skill also specifies how to handle answers that contain questions, uncertainty, or skipped items, requiring the agent to address blocking concerns before advancing to facts. Note that model invocation is disabled, so it is intended to be invoked deliberately.
A goal package under goals/<slug>/ containing working JSON files for provenance and iteration state plus authoritative human-readable markdown, built from a user interview, a reviewed fact sheet, and codebase exploration for the execution plan.
A simple, testable, verifiable description of an outcome of the goal, such as the function of a feature or specific UI/UX behavior. Each fact is flagged for whether automated verification is recommended and preselected.
It must be patient and keep waiting until the user submits, dismisses, or asks it to stop. It must never close, kill, restart, refresh, or open a second copy just because the UI looks idle; to revise, it updates the working JSON and launches a new command.
It is opt-in and suggested when the goal is vague or has many interdependent decisions, or run whenever the user asks to be grilled. For a clear, well-scoped goal it is skipped in favor of the compact interview bundle.
No. If a question can be answered by exploring the codebase, the skill explores instead of asking, and avoids obvious confirmation questions by inferring answers from the request, conversation, or shallow exploration.
Quick Setup:
.claude/skills/Repository
backnotprop/plannotator