scope - SKILL.md Agent Skill

name: scope description: "Stakeholder requirements-intake skill — helps a stakeholder or user, via a series of structured prompts, turn their ideas or requests into a right-sized, refinable scope artifact. Adaptive: asks a few high-leverage questions when a request isn't specific enough, drafts-and-trims when it's too broad. Honest about what a non-technical stakeholder cannot know. Stand-alone; does not hand off to other agents."

Scope Skill

Help a stakeholder or user, through a series of structured prompts, turn their ideas or requests into a right-sized, refinable scope artifact that a product/client-manager can clearly understand, react to, and refine — the first stage of the product work-shaping funnel.

Always announce at start: "I'm using the scope skill to shape this request."

Audience: product/client-managers — typically non-technical. Never force the stakeholder to author things they cannot know (system internals, authoritative technical scope, effort/complexity). Those belong to the refine skill.

What This Helps With

A request can land anywhere on a spectrum — sometimes not specific enough (1–2 sentences, not yet actionable), sometimes too broad (many pages spanning several distinct asks). This skill provides right-sized friction: enough structure to be actionable and honest, light enough that people actually use it.

Adaptive Intake

Detect where the request sits on the spectrum and respond accordingly:

Not specific enough → ask a capped set of high-leverage clarifying questions: at most 3–5, asked one at a time, stopping early once the request is actionable. Never an interrogation.
Too broad → skip questioning; go straight to draft-and-trim.
In between → ask only what is genuinely missing (still capped at 3–5), then draft.

Prefer multiple-choice questions where possible; open-ended is fine. Focus questions on what the stakeholder can answer: the need, the desired outcome, how they'd know it worked.

The Artifact — Bounded, Stakeholder-Honest Template

Bias short. Fill each section only as far as needed. The artifact should comfortably fit roughly one page.

Section	Cap	Purpose
Problem / Need	2–4 sentences	What is needed and why, in plain language
Desired Outcome	2–4 bullets	What "good" looks like from the stakeholder's POV — the goal, not the implementation
Boundaries (as understood)	bullets, optional	What they believe is in/out — explicitly framed as their understanding, NOT authoritative scope
Success Criteria	bullets	How they will know it is working / what they would check
Open Questions & Unknowns	bullets	Things they do not know, including explicit "needs technical input" flags
Shape	1–2 lines + optional split note	Does this feel like one cohesive need or several distinct asks, and where does it sit relative to what already exists? (NO complexity/effort — that requires a technical basis the stakeholder lacks.)

Encourage honesty about unknowns. "Open Questions & Unknowns" is a first-class section. Make it safe and normal to say "I don't know X — this needs technical input." A surfaced unknown is a success, not a gap to fill.

Deliberately excluded (these belong to the refine skill, authored by someone equipped to judge):

System Considerations / technical constraints
Authoritative In Scope / Out of Scope
Complexity / effort estimation

Right-Sizing Self-Calibration Pass

After drafting, review the artifact against these anti-patterns and trim or flag before presenting:

Too vague — a section is so thin it is not actionable. Ask one more targeted question (within the 3–5 budget) or mark it an Open Question.
Too detailed — prose has ballooned past one page, or contains implementation detail. Trim aggressively.
Forced technical detail — the stakeholder (or you) fabricated system considerations, authoritative scope, or complexity. Strip it and convert to an Open Question flagged for technical input. Do not fabricate.
Not actionable — no clear outcome or success criteria. Resolve before presenting.
Too big — does not fit one page even after trimming, or "Shape" reveals several distinct asks. Recommend splitting into multiple tickets rather than growing the document. Reserve splitting for genuinely-too-big, not hair-splitting.

Refinement Loop

Present the artifact, invite the stakeholder to react and refine, and iterate until they are satisfied. Re-run the self-calibration pass after meaningful changes.

Persistence (Environment-Adaptive)

Inline rendering is the canonical output in every environment — always render the full artifact in the conversation. That is the deliverable.

Then, only if a repository working directory is available (e.g., running in opencode), also persist it as a single evolving file:

Write/update docs/scopes/YYYY-MM-DD-<topic>.md (note the plural scopes, for consistency with docs/specs/ and docs/plans/).
Commit it so git history is the audit trail. The later refine stage edits this same file in place, so the scope→refined maturation shows up as commits to one document.
If docs/ is gitignored in the repo, keep the file as a local working artifact (still render inline).

If no repository working directory is available (e.g., a Claude organizational skill), inline is the sole output — the user copies it into their ticket/doc system. Do not attempt file mechanics.

Ticket Orientation & Next Step

The artifact is framed so it can become one ticket (or N, if split). It guides eventual implementation but is not an engineering spec — no code, no file paths, no task decomposition.

Always end the artifact with this pointer (it is a pointer, not a handoff — this skill never invokes another agent):

Next step: technical refinement with engineering (see the refine skill).

Key Principles

Right-sized friction — actionable and honest, but not onerous
Cap clarifying questions at 3–5, one at a time
Bias short; one page is the ceiling
Never force or fabricate technical detail — surface it as an unknown
Split when genuinely too big; do not balloon
Stand-alone — render inline always, persist in-repo when possible, never hand off