Arness Brainstorming

Walk the user through the entire Arness greenfield exploration pipeline in a single continuous session. The wizard invokes each pipeline skill in sequence using the Skill tool, only pausing at genuine decision gates where user input is needed. Conversational skills (discover, arch-vision) have internal feedback loops and flow without wizard-level gates.

This skill is a sequencer and decision-gate handler. It MUST NOT duplicate sub-skill logic. All pipeline work is done by the invoked skills. The wizard handles: entry routing, transitions between skills, progress display, and resumability.

Step 0: Ensure Configuration

Read ${CLAUDE_PLUGIN_ROOT}/skills/arn-spark-ensure-config/references/ensure-config.md and follow its instructions. This guarantees a user profile exists and ## Arness is configured with Arness Spark fields before proceeding.

After Step 0 completes, extract from ## Arness:

• Vision directory, Use cases directory, Prototypes directory, Spikes directory, Visual grounding directory, Reports directory

Decision Gates

The wizard pauses at exactly 9 gates. Everything else flows automatically.

Gate	When	Question	Options
G1	Entry (artifacts detected)	"It looks like you have an in-progress greenfield pipeline. Resume or start fresh?"	Resume from [detected stage] / Start fresh
G-Stress	After discover	"Run stress tests on the product concept?" then "Select which stress tests to run"	Yes/Skip; then multi-select: Interview, Competitive, Pre-Mortem, PR/FAQ
G-Name	After stress-test/concept-review	"Explore brand names for your product?"	Yes / Skip
G2	After arch-vision	"Write use cases before scaffolding?"	Write use cases (recommended) / Write use cases with expert debate (teams) / Skip to scaffold
G3	After scaffold	"Validate technical risks before visual design?"	Run spikes (recommended if risks found) / Skip to visual sketch
G4	Pre-prototype	"How should prototypes be validated?"	Standard review cycles / Expert debate (teams) for both / Expert debate (teams) for clickable only
G5	After clickable prototype	"Lock the prototype and extract features?"	Lock prototype then extract features (recommended) / Extract features without locking / Review prototype first
G6	After feature extraction	"Transition to development pipeline?"	Yes, run /arn-planning / Not yet
G7	Entry (add-feature intent + completed prototype)	"Adding a new feature. This will update use cases, iterate the prototype, and add to the feature backlog. Ready?"	Proceed / Choose which steps to include / Cancel

Workflow

Step 0.5: Extract Configuration

1. Read the project's CLAUDE.md ## Arness section (guaranteed to exist after Step 0).
2. Extract configuration fields needed for artifact detection:
   • Vision directory — for detecting product concept, architecture vision, scaffold summary, spike results, style brief, visual direction, feature backlog
   • Use cases directory — for detecting use case files
   • Prototypes directory — for detecting static and clickable prototype reports
   • Spikes directory — for detecting spike POC directories
   • Visual grounding directory — for detecting captured design references
   • Reports directory — for detecting stress test reports and concept review report

Step 1: Detect Pipeline State (Resumability)

Check which artifacts exist on disk to determine if the user is resuming a previous pipeline run. Check from most advanced to least advanced — first match wins:

Artifact	Detected State	Resume Point
User says "add feature" / "new feature" AND <prototypes-dir>/clickable/final-report.md exists	Add Feature mode	Start at Add Feature Mode (Step AF)
<vision-dir>/features/feature-backlog.md	Feature extraction complete	Resume at G6
<prototypes-dir>/locked/LOCKED.md exists AND NO <vision-dir>/features/feature-backlog.md	Prototype locked, extraction pending	Resume at feature-extract directly (invoke arn-spark-feature-extract, skip G5 lock step — already locked)
F-*.md files in <vision-dir>/features/ but NO <vision-dir>/features/feature-backlog.md	Feature extraction interrupted	Resume at feature-extract (invoke to complete/rebuild the Feature Tracker)
<prototypes-dir>/clickable/final-report.md	Clickable prototype done	Resume at G5
<prototypes-dir>/static/final-report.md	Static prototype done	Resume at clickable prototype (Step 9b)
<vision-dir>/style-brief.md	Style defined	Resume at static prototype (Step 9a)
<vision-dir>/visual-direction.md	Visual direction chosen	Resume at style-explore (Step 8)
<vision-dir>/spike-results.md	Spike complete	Resume at visual-sketch (Step 7)
<vision-dir>/scaffold-summary.md	Scaffold built	Resume at G3
UC-*.md in <use-cases-dir>/	Use cases written	Resume at scaffold (Step 5)
<vision-dir>/architecture-vision.md	Architecture defined	Resume at G2
<reports-dir>/stress-tests/concept-review-report.md AND <vision-dir>/naming-brief.md with "Final Decision" section populated	Concept review + naming complete	Resume at arch-vision (Step 3)
<reports-dir>/stress-tests/concept-review-report.md AND <vision-dir>/naming-brief.md exists but no "Final Decision" section	Naming in progress	Resume at G-Name (naming skill handles internal resume)
<reports-dir>/stress-tests/concept-review-report.md AND NO <vision-dir>/naming-brief.md	Concept review complete, naming not started	Resume at G-Name
Any stress test report (interview-report.md, competitive-report.md, premortem-report.md, prfaq-report.md) in <reports-dir>/stress-tests/ but NO concept-review-report.md	Stress tests run, review pending	Resume at concept review offer within G-Stress
<vision-dir>/product-concept.md	Product discovered	Resume at G-Stress
Greenfield fields but no artifacts	Greenfield initialized	Begin at discover (Step 2)

If artifacts are detected:

Show the detected state and ask:

Ask (using AskUserQuestion):

It looks like you have an in-progress greenfield pipeline at [detected stage]. Would you like to resume or start fresh?

1. Resume from [detected stage] — Skip to the detected resume point
2. Start fresh — Begin at Step 2 (existing artifacts are preserved)

If ambiguous or conflicting artifacts are detected (e.g., architecture-vision.md exists but no product-concept.md, or multiple conflicting states): list the detected artifacts and ask the user to clarify which state to resume from.

If no artifacts detected: skip G1 and proceed directly to Step 2.

---

Step 2: Product Discovery (Automated)

Show progress:

Spark Pipeline: DISCOVER --> stress-test --> naming --> arch-vision --> use-cases --> scaffold --> spike --> visual-sketch --> style-explore --> prototypes --> feature-extract
                ^^^^^^^^

Inform the user: "Let's start by shaping your product idea..."

Skill: arn-spark:arn-spark-discover

The discover skill has its own internal conversation loop with the product strategist agent. When it completes, product-concept.md exists in the vision directory.

---

Step 2.5: Gate G-Stress — Stress Test Decision

Show progress:

Spark Pipeline: discover --> STRESS-TEST --> naming --> arch-vision --> use-cases --> scaffold --> spike --> visual-sketch --> style-explore --> prototypes --> feature-extract
                            ^^^^^^^^^^^

Step 1 of gate — Explain and ask:

Briefly explain stress testing and each available test:

"Before committing to an architecture, you can stress-test the product concept to validate assumptions and identify blind spots. Four tests are available:

1. Synthetic User Interview — Generates synthetic personas (Pragmatist, Skeptic, Power User) from your product concept and conducts structured interviews to surface usability concerns and unmet needs.
2. Competitive Gap Analysis — Deep feature-by-feature comparison against identified competitors, revealing gaps in your positioning and missed opportunities.
3. Pre-Mortem — Assumes the product has already failed and investigates the most likely root causes, using Gary Klein's pre-mortem framework.
4. PR/FAQ — Drafts a press release for your product and then adversarially critiques it, exposing weak value propositions and messaging gaps.

Stress testing is optional but recommended — it validates WHAT you are building before committing to HOW."

Ask (using AskUserQuestion):

"Run stress tests on the product concept?"

Options:

1. Yes, run stress tests (Recommended) — Validate the concept before architecture
2. Skip to architecture vision — Proceed without stress testing

If Skip → proceed to Step 3 (arch-vision).

Step 2 of gate — Multi-select:

Ask (using AskUserQuestion with multiSelect: true):

"Select which stress tests to run (select all that apply):"

Options:

1. Synthetic User Interview — Persona-driven interviews revealing usability concerns
2. Competitive Gap Analysis — Feature comparison against competitors
3. Pre-Mortem — Investigate likely causes of product failure
4. PR/FAQ — Draft and adversarially critique a press release

Invoke each selected stress test skill in sequence:

• If Interview selected → Skill: arn-spark:arn-spark-stress-interview
• If Competitive selected → Skill: arn-spark:arn-spark-stress-competitive
• If Pre-Mortem selected → Skill: arn-spark:arn-spark-stress-premortem
• If PR/FAQ selected → Skill: arn-spark:arn-spark-stress-prfaq

After all selected tests complete:

"Stress tests complete. [N] report(s) generated. Run concept review to synthesize findings into product concept updates?"

Ask (using AskUserQuestion):

"Review and update the product concept based on stress test findings?"

Options:

1. Yes, review and update concept (Recommended) — Synthesize findings and propose concept changes
2. Skip — Keep concept as-is and proceed to architecture vision

If Yes → Skill: arn-spark:arn-spark-concept-review. After concept review completes, proceed to Step 2.7 (G-Name).

If Skip → proceed to Step 2.7 (G-Name).

---

Step 2.7: Gate G-Name — Brand Naming Decision

Show progress:

Spark Pipeline: discover --> stress-test --> NAMING --> arch-vision --> use-cases --> scaffold --> spike --> visual-sketch --> style-explore --> prototypes --> feature-extract
                                             ^^^^^^

Ask (using AskUserQuestion):

"Would you like to explore brand names for your product?"

Options:

1. Yes, find a name (Recommended) — Run the structured 4-step naming process
2. Skip — I already have a name or will name it later

If Yes → Skill: arn-spark:arn-spark-naming. After naming completes, proceed to Step 3 (arch-vision).

If Skip → proceed to Step 3 (arch-vision).

---

Step 3: Architecture Vision (Automated)

Show progress:

Spark Pipeline: discover --> stress-test --> naming --> ARCH-VISION --> use-cases --> scaffold --> spike --> visual-sketch --> style-explore --> prototypes --> feature-extract
                                                       ^^^^^^^^^^^

Inform the user: "Product concept is ready. Now let's explore the architecture..."

Skill: arn-spark:arn-spark-arch-vision

The arch-vision skill has its own internal conversation loop with the tech evaluator agent. When it completes, architecture-vision.md exists in the vision directory.

---

Step 4: Gate G2 — Use Cases Decision

Show progress:

Spark Pipeline: discover --> stress-test --> naming --> arch-vision --> USE-CASES --> scaffold --> spike --> visual-sketch --> style-explore --> prototypes --> feature-extract
                                                                                       ^^^^^^^^^

Ask (using AskUserQuestion):

"Architecture vision is complete. Write use cases before scaffolding?"

Options:

1. Write use cases (Recommended) — Structured behavioral requirements in Cockburn format
2. Write use cases with expert debate (Teams) — Multi-agent debate format with product strategist and UX specialist
3. Skip to scaffold — Proceed without formal use cases

Based on the user's choice:

• Write use cases → Skill: arn-spark:arn-spark-use-cases
• Expert debate → Skill: arn-spark:arn-spark-use-cases-teams
• Skip → mark use-cases as skipped and proceed to Step 5

---

Step 5: Scaffold (Automated)

Show progress:

Spark Pipeline: discover --> stress-test --> naming --> arch-vision --> use-cases --> SCAFFOLD --> spike --> visual-sketch --> style-explore --> prototypes --> feature-extract
                                                                                                 ^^^^^^^^

Inform the user: "Setting up the project skeleton with the chosen tech stack..."

Skill: arn-spark:arn-spark-scaffold

The scaffold skill creates the project structure, installs dependencies, and produces a buildable project. When it completes, scaffold-summary.md exists in the vision directory.

---

Step 6: Gate G3 — Spike Decision

Show progress:

Spark Pipeline: discover --> stress-test --> naming --> arch-vision --> use-cases --> scaffold --> SPIKE --> visual-sketch --> style-explore --> prototypes --> feature-extract
                                                                                                         ^^^^^

Before asking, read architecture-vision.md from the vision directory and check for a "Risks" or "Known Risks" section. Count the number of identified risks.

Ask (using AskUserQuestion):

"Scaffold complete. Validate technical risks before moving to visual design?"

Options (if risks were found in architecture vision):

1. Run spikes (Recommended — [N] risks identified in architecture vision) — Validate risky assumptions with proof-of-concept code
2. Skip to visual sketch — Proceed to visual design without risk validation

Options (if no risks section found):

1. Run spikes anyway — Validate key assumptions even though no explicit risks were documented
2. Skip to visual sketch (Recommended) — Proceed to visual design

If Run spikes:

Skill: arn-spark:arn-spark-spike

If Skip: mark spike as skipped and proceed to Step 7.

---

Step 7: Visual Sketch (Automated)

Show progress:

Spark Pipeline: discover --> stress-test --> naming --> arch-vision --> use-cases --> scaffold --> spike --> VISUAL-SKETCH --> style-explore --> prototypes --> feature-extract
                                                                                                          ^^^^^^^^^^^^^

Inform the user: "Generating visual direction proposals..."

Skill: arn-spark:arn-spark-visual-sketch

The visual-sketch skill generates multiple direction proposals as live pages on the dev server. When it completes, visual-direction.md exists in the vision directory.

---

Step 8: Style Explore (Automated)

Show progress:

Spark Pipeline: discover --> stress-test --> naming --> arch-vision --> use-cases --> scaffold --> spike --> visual-sketch --> STYLE-EXPLORE --> prototypes --> feature-extract
                                                                                                                            ^^^^^^^^^^^^^

Inform the user: "Defining the design system from the chosen visual direction..."

Skill: arn-spark:arn-spark-style-explore

The style-explore skill defines color palettes, typography, component styles, and toolkit configuration. When it completes, style-brief.md exists in the vision directory.

---

Step 9: Gate G4 — Prototype Validation Mode

Show progress:

Spark Pipeline: discover --> stress-test --> naming --> arch-vision --> use-cases --> scaffold --> spike --> visual-sketch --> style-explore --> PROTOTYPES --> feature-extract
                                                                                                                                          ^^^^^^^^^^

Ask (using AskUserQuestion):

"Style system defined. How should prototypes be validated?"

Options:

1. Standard review cycles — Build-review-iterate with expert scoring and UX judge verdict
2. Expert debate (teams) for both — Multi-agent debate for static and clickable prototypes
3. Expert debate (teams) for clickable only — Standard for static, debate for clickable

Based on choice, invoke the prototype skills in sequence:

Standard:

Skill: arn-spark:arn-spark-static-prototype Then: Skill: arn-spark:arn-spark-clickable-prototype

Teams for both:

Skill: arn-spark:arn-spark-static-prototype-teams Then: Skill: arn-spark:arn-spark-clickable-prototype-teams

Teams for clickable only:

Skill: arn-spark:arn-spark-static-prototype Then: Skill: arn-spark:arn-spark-clickable-prototype-teams

After each prototype skill completes, show an intermediate progress update before invoking the next.

---

Step 10: Gate G5 — Post-Prototype Actions

Show progress:

Spark Pipeline: discover --> stress-test --> naming --> arch-vision --> use-cases --> scaffold --> spike --> visual-sketch --> style-explore --> prototypes --> FEATURE-EXTRACT
                                                                                                                                                            ^^^^^^^^^^^^^^^

Before asking, check if <prototypes-dir>/locked/LOCKED.md exists.

If prototype is already locked:

"Prototype is already locked (see prototypes/locked/LOCKED.md). Proceeding directly to feature extraction."

Skill: arn-spark:arn-spark-feature-extract

Skip the G5 question — locking is already done. After feature-extract completes, proceed to G6.

If prototype is NOT locked:

Ask (using AskUserQuestion):

"Prototype validation complete. What next?"

Options:

1. Lock prototype and extract features (Recommended) — Protect the validated prototype, then build the feature backlog
2. Extract features without locking — Build the backlog without prototype protection
3. Review prototype first — Look at the validation results before deciding

If Lock and extract:

Skill: arn-spark:arn-spark-prototype-lock Then: Skill: arn-spark:arn-spark-feature-extract

If Extract without locking:

Skill: arn-spark:arn-spark-feature-extract

If Review first:

• Read the final reports from both static and clickable prototype directories
• Present a summary of the validation results (verdicts, scores, key findings)
• Then re-ask: "Lock and extract, or extract only?"

---

Step 11: Gate G6 — Transition to Development

Show progress:

Spark Pipeline: discover --> stress-test --> naming --> arch-vision --> use-cases --> scaffold --> spike --> visual-sketch --> style-explore --> prototypes --> feature-extract
                       ✓            ✓              ✓              ✓             ✓           ✓           ✓                ✓               ✓               ✓

Before asking about transition, check if features have been uploaded to the issue tracker:

1. Read ## Arness config for Issue tracker field (github, jira, or none). If not present, treat as none.

2. If Issue tracker is none: skip this check.

3. If Issue tracker is github or jira: a. Read <vision-dir>/features/feature-backlog.md and check the Feature Tracker table's Issue column. b. If ALL features have -- in the Issue column (none uploaded):

   Ask (using AskUserQuestion):

   Your feature backlog has [N] features but none have been uploaded to [GitHub/Jira] yet. Would you like to upload them now? This creates issues for team visibility and enables /arn-code-pick-issue for feature selection.

   1. Upload now — Create issues for all features
   2. Skip — Continue without uploading

   If Upload now → Skill: arn-spark:arn-spark-feature-extract (the skill detects existing features and runs Step 7 upload) If Skip → Continue to transition question

   c. If SOME features have issue numbers (partial upload):

   Ask (using AskUserQuestion):

   [M] of [N] features have been uploaded to [GitHub/Jira]. Would you like to upload the remaining [K]?

   1. Upload remaining — Create issues for the remaining features
   2. Skip — Continue without uploading

   If Upload remaining → Skill: arn-spark:arn-spark-feature-extract If Skip → Continue to transition question

   d. If ALL features have issue numbers: skip this check silently.

Then proceed:

Ask (using AskUserQuestion):

"Greenfield exploration is complete. Feature backlog is ready. Transition to the development pipeline?"

Options:

1. Yes, start developing — Run /arn-planning to begin planning features from the backlog. Arness auto-configures code patterns, plans, specs, and report templates on first use.
2. Not yet — Stay in greenfield mode; transition later

If Yes:

Skill: arn-code:arn-planning

If Not yet:

• Show what was completed
• Inform: "Run /arn-planning when ready to transition. Arness auto-configures on first use. Your greenfield artifacts are preserved."

---

Step 12: Complete

Show final progress:

Spark Pipeline: discover --> stress-test --> naming --> arch-vision --> use-cases --> scaffold --> spike --> visual-sketch --> style-explore --> prototypes --> feature-extract
                       ✓            ✓              ✓              ✓             ✓           ✓           ✓                ✓               ✓               ✓

Present a completion summary:

• Product concept: <vision-dir>/product-concept.md
• Brand name: <vision-dir>/naming-brief.md — [chosen name] (or "skipped")
• Architecture: <vision-dir>/architecture-vision.md
• Use cases: [count] use cases in <use-cases-dir>/ (or "skipped")
• Scaffold: <vision-dir>/scaffold-summary.md
• Spikes: <vision-dir>/spike-results.md (or "skipped")
• Visual direction: <vision-dir>/visual-direction.md
• Style: <vision-dir>/style-brief.md
• Static prototype: <prototypes-dir>/static/final-report.md
• Clickable prototype: <prototypes-dir>/clickable/final-report.md
• Prototype lock: [yes / no]
• Feature backlog: <vision-dir>/features/feature-backlog.md — [feature count] features
• Development transition: [initialized / deferred]

"Pipeline complete. Run /arn-brainstorming again to re-explore, or /arn-planning for the development pipeline."

---

Add Feature Mode

Triggered when the user wants to add a new feature to a completed greenfield project. This mode sequences the incremental update skills in dependency order: use cases provide the behavioral foundation, the prototype validates the UI, and feature extraction adds to the backlog. Handles locked prototypes by iterating to a new version and re-locking.

Read ${CLAUDE_PLUGIN_ROOT}/skills/arn-brainstorming/references/add-feature-flow.md for the full Add Feature mode workflow.

---

Error Handling

• Sub-skill fails or errors out: Present the error to the user. Ask (using AskUserQuestion): retry this step / skip this step / abort the wizard. If retry, re-invoke the same skill. If skip, continue to the next gate. If abort, show what was completed and exit.
• User says "stop" or "pause": Show what has been completed and what's next. Inform the user they can resume by running /arn-brainstorming again (artifact detection will pick up where they left off).
• Arness greenfield not initialized: Handled by Step 0 ensure-config, which auto-configures with sensible defaults.
• Architecture vision has no risks section at G3: Adjust the spike recommendation — default suggestion becomes "Skip" instead of "Run spikes". Show: "No explicit risks identified in the architecture vision."
• Scaffold fails to build: The scaffold skill has its own error handling with retries. If it ultimately fails, ask: retry scaffold / abort wizard. Do not skip scaffold — it is required for all subsequent visual steps.
• Prototype validation does not reach PASS verdict: After the sub-skill returns, inform the user: "The prototype did not reach a PASS verdict. Continue to feature extraction anyway, or iterate more on the prototype?" If iterate: re-invoke the prototype skill. If continue: proceed to G5.
• arn-spark plugin not installed: If skill invocations fail because the greenfield plugin is missing, inform: "The arn-spark plugin must be installed. Run /plugin install arn-spark@arn-marketplace."
• Teams variant not available at G2 or G4: If the user selects a teams variant but the environment does not support Agent Teams (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1), inform them and suggest the standard option instead.
• Product concept already exists when starting fresh: The discover skill handles this internally (asks to replace or refine). The wizard does not need separate handling.
• Add Feature mode invoked without completed prototype: Redirect to normal wizard flow. Inform: "Add Feature mode requires a completed clickable prototype. Let's start with the full pipeline instead."
• Locked prototype during Add Feature: The lock protects the frozen snapshot, not the working directory. Prototype iteration creates a new version alongside the locked one. After iteration, the wizard offers to re-lock with the new version.
• Feature extraction interrupted (partial feature files exist without Feature Tracker): Detected by finding F-*.md files in the features directory without feature-backlog.md. Invoke arn-spark-feature-extract which handles resume and rebuilds the Feature Tracker from existing feature files.
• Features not uploaded to issue tracker: Detected at G6 by checking the Feature Tracker's Issue column. If all entries show --, offer to upload before transitioning to development.

Constraints

• This skill MUST NOT duplicate sub-skill logic. It only handles sequencing and decision gates.
• All pipeline work is done by the invoked skills via the Skill tool.
• Progress display uses the compact format shown above — one line showing the full pipeline with the current stage highlighted.
• The wizard runs in normal conversation (not plan mode).
• Skill invocations use the Skill tool with fully qualified names: Skill: arn-spark:arn-spark-<name> for greenfield skills, Skill: arn-code:arn-planning for the development pipeline transition.
• Add Feature mode uses the same sub-skill invocation pattern as the main pipeline.
• Add Feature mode never unlocks or modifies a locked prototype snapshot. It iterates to a new version and optionally re-locks.