spectre-scope

name	spectre-scope
description	👻 | Interactively scope a feature or improvement, generating a complete Scope document with IN, OUT, and ANTI-SCOPE boundaries -- primary agent
user-invocable	true

scope

Input Handling

Treat the current command arguments as this workflow's input. When invoked from a slash command, use the forwarded $ARGUMENTS value.

scope: Interactive Feature Scoping

Collaborative workflow for structuring unstructured thoughts into clear scope boundaries through a grounded hypothesis and targeted clarifications. Focuses on user value and scope boundaries before technical considerations. Output: scope document with clear boundaries (IN / OUT / ANTI-SCOPE), user value, load-bearing assumptions, and key decisions saved to {OUT_DIR}/concepts/scope.md.

ARGUMENTS

$ARGUMENTS

Step 1: Immediate Reply & Context Pickup

• Action — ImmediateReply: Respond before any tools.

  • If FROM_KICKOFF=true → acknowledge kickoff context, read KICKOFF_DOC, extract (Core Problem, User Value, Decisions Made, Remaining Ambiguities, Key Code Refs), SKIP to Step 4
  • Else If ARGUMENTS empty → probe for context enthusiastically and WAIT for user
  • Else → proceed with the steps below
  • CRITICAL: No tool calls in this step except reading the kickoff doc when FROM_KICKOFF=true.

• Action — DetectExistingArtifacts: Before proposing a hypothesis, check for prior context that may already answer questions.

  • branch_name=$(git rev-parse --abbrev-ref HEAD 2>/dev/null || echo unknown)
  • Look for: docs/tasks/{branch_name}/concepts/scope.md, docs/tasks/{branch_name}/task_context.md
  • If prior scope.md exists → read fully; treat this invocation as a re-scope. Surface what was already settled and ask only about what's new or changed.
  • If captured project knowledge is available (via spectre-apply-style memory or session logs) → surface decisions/patterns/gotchas relevant to this area before asking the user.

Step 2: Ground the Hypothesis (Fast)

SKIP IF FROM_KICKOFF=true — kickoff already grounded.

• Action — FastGround: Run ONE fast lookup to ground the hypothesis in repo reality. Pick the most informative single call:

  • @finder for one targeted query — when you need to know what files/components exist for this area
  • One grep or glob — when you need to verify a specific assumption ("does X already exist?")

  Constraints:

  • One call only. This is grounding, not research. Tech research is spectre-plan's job.
  • Fast. If the call would take more than a few seconds, skip it and ask the user instead.
  • If you need more than one signal, that's a sign this should be a spectre-kickoff rather than a spectre-scope.

Step 3: Interactive Scope Exploration

SKIP IF FROM_KICKOFF=true

• Action — ExploreScope: Collaborative dialogue focused on boundaries, user experience, and anti-scope.

  CRITICAL: Focus on WHAT, not HOW. Defer all technical/implementation questions to spectre-plan. Only ask technical questions if the scope itself is inherently technical (e.g., "migrate database from X to Y").

  PATTERN: Lead with a grounded hypothesis (informed by Step 2's lookup) AND ask 5–8 questions across multiple dimensions. Mark each question as (blocking) or (optional) so the user can skim and skip.

  FIRST RESPONSE FORMAT:

  Here's my read on this, grounded in {one-line of what Step 2 surfaced}:

  Hypothesis: [problem statement, who it affects, proposed IN / OUT / ANTI-SCOPE]

  Questions:

  1. (blocking) [User problem question]
  2. (blocking) [Boundary edge question]
  3. (blocking) [Anti-scope question]
  4. (optional) [UX flow question]
  5. (optional) [Alternative approach question]
  6. (optional) [Edge case question]
  7. (optional) [Success criteria question]

  Answer the blocking ones; skip optional if obvious.

  Question types (mix from these; lean on blocking for boundaries and anti-scope):

  • User & Problem: Who feels this most? What triggers the need? What's the cost of not solving it?
  • UX & Feel: Should this feel fast or thorough? Guided or flexible? What's the ideal flow?
  • Boundaries (IN/OUT): What about [adjacent thing]—IN or OUT? What's essential for v1?
  • Anti-Scope: What problem are we intentionally not solving here? (Different from OUT — anti-scope clarifies the philosophical edge of the feature, not just deferred work.)
  • Alternatives: I could see this as [A] or [B]. Which direction?
  • Edge cases: What happens when [unusual situation]?
  • Success: What makes you say "this shipped well"?

  DO NOT ask about: implementation approach, technical trade-offs, architecture, or integration details — those belong in spectre-plan.

• Action — IterateBoundaries: After user responds, refine boundaries and ask targeted follow-ups on gaps.

  Current Scope: ✅ IN: [list] ❌ OUT: [list] 🚫 ANTI-SCOPE: [problems we're intentionally not solving] ⚠️ Unsure: [edge cases]

  Any items to move? Add exclusions? Clarify edges? Reply 'looks good' to continue.

• Wait — User confirms scope boundaries are accurate.

Step 4: Targeted Clarifications

• Action — DetermineOutputDir:

  • If FROM_KICKOFF → use same dir as kickoff doc
  • Else → OUT_DIR = user_specified || docs/tasks/{branch_name}
  • mkdir -p "$OUT_DIR/concepts" "$OUT_DIR/clarifications"

• Action — DetectClarificationMethod: Check if AskUserQuestion tool is available.

  • If available → use inline clarification flow (Path A)
  • If not available → use file-based clarification flow (Path B)

Path A: Inline Clarifications (AskUserQuestion available)

• Action — AskInline: Generate 3–6 questions based ONLY on remaining scope ambiguities from Step 3 (or kickoff's "Remaining Ambiguities").

  • Ask the most critical first (up to 4 at a time, the tool limit).
  • For trade-off decisions, present options with concise pros/cons.
  • If more than 4 questions, ask in batches — most important first.

  CRITICAL: Only ask about unresolved scope ambiguities. Technical questions belong in spectre-plan, not here.

Path B: File-Based Clarifications (no AskUserQuestion)

• Action — GenerateTargetedQuestions: Create 3–6 questions in {OUT_DIR}/clarifications/scope_clarifications_{YYYY-MM-DD_HHMMSS}.md:

  • Header: concept name, confirmed boundaries so far
  • Each question with a <response></response> block

• Action — PromptUser: "Saved clarifications to {path}. Answer in <response> blocks. Reply 'Read it' when ready."

• Wait — User replies.

• Action — ReadClarifications: Read the file; use responses (proceed with intelligent assumptions if blocks are empty).

Step 5: Create Scope Document

• Action — CreateScopeDoc: Generate {OUT_DIR}/concepts/scope.md (use scoped filename if one exists).

  Priority: User value and boundaries BEFORE technical details.

  Sections:

  1. The Problem — pain, impact, current state
  2. Target Users — primary, secondary, needs
  3. Success Criteria — outcomes; prefer measurable assertions over prose
  4. User Experience — journeys, principles, trade-offs
  5. Scope Boundaries — IN / OUT / ANTI-SCOPE / Maybe / Future
  6. Load-Bearing Assumptions — assumptions that, if wrong, invalidate this scope. Each assumption gets a short "if this is false, …" consequence so future-you can recognize when to come back.
  7. Constraints — platform, perf, a11y, scale (user-provided only)
  8. Decisions — from clarifications + rationale
  9. Risks — UX, scope creep, open questions
  10. Next Steps — spectre-plan or spectre-create_tasks, complexity S/M/L

  Note on ANTI-SCOPE vs OUT:

  • OUT = we're not building it (yet, or for this release).
  • ANTI-SCOPE = we're intentionally not solving this problem — it clarifies the philosophical edge of what this feature is for.

Step 6: Final Review & Next Steps

• Action — PresentDocForReview: Show final boundaries and next steps together.

  Scope Complete: {OUT_DIR}/concepts/scope.md

  Final Boundaries: ✅ IN: [from doc] ❌ OUT: [from doc] 🚫 ANTI-SCOPE: [from doc] ⚠️ Maybe/Future: [from doc]

  Load-Bearing Assumptions (top 1–3 — if any prove false, return to scope): [from doc]

  Docs saved: {OUT_DIR}/concepts/scope.md, clarifications under {OUT_DIR}/clarifications/

  Reply with edits — otherwise pick a next step:

• Action — RenderFooter: Render Next Steps using Skill(spectre-guide) skill.

  NOTE: Do NOT wait for explicit approval. Present next steps inline with the review. The user can reply with scope edits OR jump straight into a next step command. If they reply with edits, apply them to the scope doc and re-present.