product-explore

name	product-explore
description	Turn a raw product idea into a validated proposal with a recommended scope decision.
when_to_use	Use when going from rough idea to written proposal. Also when user says "should I build this", "validate this idea", "is this a real product", "explore this product idea", "competitor scan", "scope this", or supplies a half-baked concept they want pressure-tested.
user-invocable	true
argument-hint	[idea or flags: --auto/--fast/--quick/--no-collision/--update/--branch/--deep-dive/--export]
allowed-tools	Agent AskUserQuestion Read Write Edit Glob Grep Bash Skill

/product-explore

Product exploration pipeline that turns raw ideas into validated proposals. Acts as a strategic consultant — interprets data, recommends actions, and proactively suggests next steps.

Phase 1 combines /clarify (requirements depth) with a 6-question demand probe (product-market validation). Phase 6 runs a CEO-level scope review on the drafted proposal — EXPANSION / SELECTIVE EXPANSION / HOLD SCOPE / REDUCTION — before writing the final recommendation. Every session ends with "The Assignment," a concrete real-world action the founder commits to before next session.

Architecture:

• This skill: entry point, preflight checks, flag parsing, mode routing
• product-explorer agent: thin orchestrator (mode dispatch, checkpoints, state)
• product-researcher agent: Phase 2 research (facts + hypotheses)
• product-shaper agent: Phase 4 creative (vision, naming, canvas, positioning)
• product-stress-tester agent: Phase 5 adversarial (assumptions, experiments)
• product-collision-analyst agent: Phase 5.5 cross-referencing (Opus)

Flag Parsing

Parse flags from $ARGUMENTS before mode detection. Flags are combinable.

flags = {
  auto: $ARGUMENTS contains "--auto",
  quick: $ARGUMENTS contains "--quick",
  no_collision: $ARGUMENTS contains "--no-collision"
}
# Strip flags from $ARGUMENTS before mode detection
idea_text = $ARGUMENTS with all --flags removed

Flag Conflict Validation

After parsing, check for conflicts. Report and STOP if invalid:

Conflict	Why	Resolution
--fast + --quick	Both modify pipeline depth, incompatible	"Use --fast for 3-phase gut check, or --quick for full pipeline without stress test. Not both."
--no-collision + --update	Update mode's value IS re-running collision	"Update mode always runs collision analysis. Use --update without --no-collision."
--fast + --no-collision	Fast mode never runs collision anyway	Silently ignore --no-collision (no error, just redundant)
--quick + --no-collision	Valid combo	Allow — skip stress test AND collision

Mode Detection

After stripping flags, detect mode from remaining arguments:

Argument	Mode	Action
--fast or --fast [idea]	Fast	3-phase gut check (~50K tokens)
--update	Update	Feed experiment results into existing proposal
--branch "description"	Branch	Fork exploration into alternative direction
--deep-dive "topic"	Deep Dive	Focused research on one topic
--export pitch	Export	Reformat proposal as pitch deck outline
(anything else)	Deep (default)	Full adaptive pipeline (~150K tokens)

Common Flag Combos

Combo	Use Case	Est. Tokens
--auto	Rich brief, full depth, minimal pauses	~130K
--auto --fast	Weekly screening (3-4 ideas)	~40K
--auto --quick	Fast with collision insights	~90K
--auto --quick --no-collision	Fastest full proposal	~70K
--quick	Interactive but skip stress test	~110K

Phase 0: Preflight Check

Before spawning the agent, validate ALL dependencies. Run ALL checks before reporting — never fail one-at-a-time.

Required Plugins

Check that these plugin directories exist in the cache:

pm-skills (6 sub-plugins):

~/.claude/plugins/cache/pm-skills/

Provides: user-personas, market-sizing, competitor-analysis, job-stories, pre-mortem, beachhead-segment, product-vision, value-proposition, lean-canvas, identify-assumptions-new, prioritize-assumptions, brainstorm-experiments-new, positioning-ideas

marketingskills:

~/.claude/plugins/cache/marketingskills/

Provides: customer-research

nextc-claude (bundled):

• clarify — always available since it's bundled with this plugin

How to Check

ls ~/.claude/plugins/cache/pm-skills/ 2>/dev/null
ls ~/.claude/plugins/cache/marketingskills/ 2>/dev/null

Optional Checks

• nextc-ecc: Check ~/.claude/plugins/cache/*/nextc-ecc/ (glob — the plugin nests under its marketplace name). Not used by product-explore directly, but the next step (/flutter-kickoff or /feature-dev) requires it. Warn early.

On Failure

Report ALL missing items at once with exact install commands:

Preflight FAILED. Missing dependencies:

PLUGINS:
  - pm-skills — Install:
    /plugin marketplace add phuryn/pm-skills
    /plugin install pm-market-research@pm-skills
    /plugin install pm-execution@pm-skills
    /plugin install pm-go-to-market@pm-skills
    /plugin install pm-product-strategy@pm-skills
    /plugin install pm-product-discovery@pm-skills
    /plugin install pm-marketing-growth@pm-skills

  - marketingskills — Install:
    /plugin marketplace add coreyhaines31/marketingskills
    /plugin install marketing-skills@marketingskills

OPTIONAL:
  - nextc-ecc not detected — your next step (/flutter-kickoff or /feature-dev) needs it.
    /plugin install nextc-ecc@nextc-claude

Install missing dependencies and run /product-explore again.

STOP here. Do NOT proceed to the agent until all required plugins are installed.

On Success

Preflight passed.
  Plugins: pm-skills ✓, marketingskills ✓
  Optional: nextc-ecc [✓/missing]
  Skills: [N] available
  Flags: auto=[yes/no], quick=[yes/no], no-collision=[yes/no]

Cost & Time Estimate

After preflight passes, show the estimated cost and time for the selected mode + flags:

Estimated run:
  Mode: [deep/fast/update/branch/deep-dive/export]
  Flags: [auto, quick, no-collision — or none]
  Tokens: ~[N]K
  Time: ~[N] min
  Opus calls: [N] (collision analysis[, vision synthesis])

Mode + Flags	Tokens	Time	Opus Calls
deep	~150K	~20 min	2 (collision + vision)
deep --auto	~130K	~15 min	2
deep --quick	~110K	~15 min	1 (collision only)
deep --auto --quick	~90K	~12 min	1
deep --auto --quick --no-collision	~70K	~10 min	0
fast	~50K	~8 min	0
fast --auto	~40K	~6 min	0
update	~60K	~10 min	1 (collision)
branch	~80K	~12 min	1 (collision)
deep-dive	~20K	~5 min	0
export	~10K	~3 min	0

Spawn the Agent

After showing the estimate, spawn the product-explorer orchestrator:

Agent(
  subagent_type: "nextc-product:product-explorer",
  prompt: """
  Mode: [deep/fast/update/branch/deep-dive/export]
  Flags: auto=[yes/no], quick=[yes/no], no-collision=[yes/no]
  Arguments: $ARGUMENTS (with flags stripped)
  Working directory: [current project root]
  Template path: ~/.claude/plugins/cache/nextc-claude/nextc-product/skills/product-explore/templates/proposal.md

  [For update/branch/deep-dive/export modes:]
  Existing exploration: docs/explore/ exists — read it for prior state.

  [For deep/fast modes:]
  User's idea: [the idea text from $ARGUMENTS]
  """
)

Mode-Specific Preconditions

Mode	Precondition	On Failure
--update	docs/explore/ must exist	"No existing exploration found. Run /product-explore first."
--branch	docs/explore/ must exist	"No existing exploration found. Run /product-explore first."
--deep-dive	docs/explore/ must exist	"No existing exploration found. Run /product-explore first."
--export pitch	docs/proposal.md must exist	"No proposal found. Run /product-explore first."
--fast	None	Proceed
(default)	None	Proceed

Post-Run Handoff

After the agent writes the proposal, surface a single next-step recommendation based on what the user committed to in "The Assignment." Common handoffs:

Proposal state	Next step to offer
Working product name is provisional / placeholder	/aso-naming --from-proposal — runs the ASO naming framework (Brand+Keyword formula, 4 angles, 12-point scoring, collision + WoM checks). Spec at specs/aso-pipeline/strategic-product-naming.md. ~5 min.
Name is locked AND user is ready to build	/flutter-kickoff or /unity-kickoff
Name is locked AND store launch is the next milestone	/aso-pipeline build (the build step in /aso-pipeline will itself offer /aso-naming if the title is still provisional in the brief)
Major assumptions still untested	The Assignment itself — stop layering tools on top of an idea that needs a real-world check first

Lead with one recommendation, not a menu. The product-shaper agent already handled vision-level naming during Phase 4; /aso-naming is the ASO-grade discoverability pass that the user can run before committing to the App Store listing or trademark filing.