From project-tools
Identify ambiguous or missing specification details and resolve them through targeted questions before implementation.
How this skill is triggered — by the user, by Claude, or both
Slash command
/project-tools:beepus-maximus-project-clarify-specThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
This is the canonical workflow. Claude slash commands may call this skill, but they must not duplicate its process. Codex should use this skill directly. Project-specific documents, scripts, tools, and paths named below are optional context; read or run them only when they exist in the current repository.
This is the canonical workflow. Claude slash commands may call this skill, but they must not duplicate its process. Codex should use this skill directly. Project-specific documents, scripts, tools, and paths named below are optional context; read or run them only when they exist in the current repository.
Use this skill as the Clarify Spec workflow. Your job is to systematically identify underspecified, ambiguous, or missing areas in a feature specification and resolve them through targeted questions — before implementation begins.
This skill modifies specifications only. It never writes implementation code.
implement skill on a new or complex specproject-scope skill flags a spec as needing work before implementationdocs-audit skill reports underspecification findingsBefore starting, read ALL of:
If the user provided a feature name, find the matching spec in the project specifications or requirements documents.
If ambiguous, list candidates and ask. Do not guess.
Read the spec fully before proceeding.
Scan the spec against each concern area. For each area, classify as Complete, Partial, or Missing.
| Concern Area | What to Check |
|---|---|
| Functional Scope | Are all user-visible behaviours described? Is the boundary between in-scope and out-of-scope clear? |
| Domain & Data Model | Are all entities and types defined? Are relationships and cardinality explicit? Are persistence implications stated? |
| Interaction & UX | Are entry points, flows, and exit points described? Are state transitions (loading, empty, error, success) covered? |
| Business Rules | Are constraints and invariants explicit? Are they testable? Do any rules contradict each other? |
| Edge Cases | Are boundary conditions covered? What happens with empty data, max data, concurrent access, offline state? |
| Dependencies | Are spec dependencies listed? Are they implemented (check current-scope.md)? Are integration points described? |
| Non-Functional | Performance expectations? Data volume assumptions? Background behaviour? Permission requirements? |
| Terminology | Are all domain terms in the glossary? Could any term be interpreted differently by a different reader? |
| Acceptance Criteria | Does every user scenario have at least one criterion? Is every criterion testable with a concrete input/output? |
| Completion Signals | How do we know the feature is done? Are success criteria measurable? |
From the scan, collect all gaps. Prioritize by impact on implementation:
If no blocking or risky gaps exist → report "Spec is ready for implementation" and suggest implement skill. Done.
Ask questions one at a time, in priority order (blocking first, then risky). Maximum 5 questions per session.
Each question must include:
Format:
### Question N of M
**Gap:** [what is missing or ambiguous]
**Section:** [spec section this affects, or "New section needed"]
**Impact:** [Blocking | Risky]
**Why this matters:** [what goes wrong without an answer]
**Recommendation:** [your suggested answer and reasoning]
**Options:**
| Option | Description | Trade-off |
|--------|-------------|-----------|
| A | [option] | [pro/con] |
| B | [option] | [pro/con] |
| C | [recommendation — marked] | [pro/con] |
Wait for the user's answer before asking the next question.
Trigger: During Phase 3, if any concern area was classified as Missing (not Partial — entirely absent), activate this phase. Skip if all gaps are Partial or Nice-to-have.
For each Missing section:
[ASSUMED] inline.### Draft Section: [Section Name]
**Status:** This section was entirely missing. Here is a proposed draft based on [context sources].
[Draft content with [ASSUMED] markers on any assumptions]
**Assumptions made:**
1. [Assumption] — based on [evidence]
2. [Assumption] — based on [evidence]
Accept this draft? (yes / yes with changes / reject and I'll ask questions instead)
Rules for drafting:
After drafted sections are accepted (or rejected), continue to Phase 5 for any remaining questions from Phase 4.
After each answer:
After all questions are resolved:
Draft and all blocking gaps are resolved, ask if status should move to ApprovedPresent:
| Rationalization | Reality |
|---|---|
| "The spec looks fine to me" | Run the taxonomy scan. Humans miss gaps that systematic checking catches. |
| "I'll ask all 5 questions at once" | One at a time. Each answer may change the next question. |
| "I'll just fill in reasonable defaults" | Defaults are assumptions. Assumptions become bugs. Ask. |
| "This gap is obvious, I'll infer the answer" | Present the inference as a recommendation and let the user confirm. |
| "The spec needs a complete rewrite" | Clarify is surgical. If the spec needs rewriting, that is a different task. |
After clarification is complete, suggest the appropriate next action:
| Situation | Suggested Skill | Why |
|---|---|---|
| Spec is now approved and ready | implement skill | Start building |
| Want to validate spec quality further | docs-audit skill | Cross-artifact consistency check |
| Spec has drifted from existing code | sync-spec skill | Reconcile with reality first |
| Multiple specs need clarification | clarify-spec skill again | Run on the next spec |
| Gap is deferred — worth revisiting later | idea-seed skill | Capture the idea with a trigger condition so it surfaces when relevant work begins |
| Artifact state is confusing or unclear | forensics skill | Investigate what happened before trying to clarify |
NEEDS CLARIFICATION in the spec and move onGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub 4eleven7/claude-skills --plugin project-tools