customer-discovery

Pack Availability Guard

Before telling the user to run a skill from another project-local pack, check .agents/project.json.enabled_packs. If the target pack is not enabled, recommend npx skillpacks install <pack> from the project shell, instead of the target skill. After install, tell Codex users to start a fresh Codex CLI session if the $ skill list remains stale. Global skills are always valid. Skills from this same pack are valid because the current skill is already running from that pack.

Customer Discovery — Orchestrator

Invoke as $customer-discovery.

This is an orchestrator skill using the parent router delegation pattern. It detects context, bootstraps ICP candidates, recommends applicable customer-discovery frameworks, and synthesizes their outputs into unified ICP research. Individual frameworks live as child skills under frameworks/.

Available frameworks:

Framework	Slug	Creator	Lens
W3 Hypothesis	w3-hypothesis	Schwartzfarb	WHO/WHAT/WHY hypothesis generation + disproval
JTBD Needs Analysis	jtbd-needs	Ulwick / Christensen	Needs-based segmentation via jobs and outcomes
Four Forces	four-forces	Moesta	Push/Pull/Anxiety/Habit switching moment analysis
Five Rings of Buying Insight	five-rings	Revella	Decision psychology and buyer journey
Seven Dimensions	seven-dimensions	Lincoln Murphy	Ready/Willing/Able/Success/Acquisition/Ascension/Advocacy scoring
PMF Engine	pmf-engine	Vohra / Supan	Empirical ICP from existing user data (Sean Ellis test + HXC)

Report-First Approval Gate

Default to scope-first approval: before synthesized research, inspect only enough repository, user, and source context to propose research scope, source plan, assumptions, output paths, and approval questions in a review alignment page plus a concise conversation summary.

Do not perform synthesized research, rank candidates, make recommendations, or write working packets or canonical deliverables until final compiled YAML approves the research scope. Minimal pre-approval discovery may identify available files, source categories, and open questions; label it as scope evidence, not findings.

After approved research-scope YAML, perform the research and write only the non-canonical working packet defined in the staged workflow. Then update the review alignment page with findings and stop again for feedback-only YAML or final compiled YAML artifact approval before creating or updating canonical research, spec, or task files.

Do not include downstream or cross-skill command recommendations while a scope, framework findings, or synthesis approval is pending. The approval request itself is the next action. Parent-loop continuation is not downstream routing: after framework completion, hand control back to $customer-discovery, never an execution-loop command or a path-shaped child framework invocation. Only emit downstream next-skill routing after the synthesized icp.md artifact has been approved and written.

Staged Research Workflow

Use this staged workflow for synthesized research or report outputs that would create or update canonical research, spec, or task files.

1. Stage 1 - Scope discovery and approval. Inspect only enough repository, user, and source context to propose research scope, source plan, assumptions, output paths, and approval questions. Build the review HTML alignment page before synthesized research. The page must render the proposed scope, available source categories, known context, assumptions/confidence, proposed working-packet and canonical output paths, and research-scope approval gates. Stop for final compiled YAML approval of the research scope. Do not perform synthesized research, rank candidates, make recommendations, or write working packets, canonical research, spec, or task files in Stage 1.
2. Stage 2 - Research and artifact review. Only after approved research-scope YAML with no unresolved needs-clarification, unresolved down feedback, or other unresolved negative feedback, perform the synthesized research, run required source/code checks, and write only a non-canonical working packet: flat mode uses research/_working/preliminary-<skill>-research.md; product-path mode uses research/{slug}/_working/preliminary-<skill>-research.md. Replace <skill> with this skill's name value. Raw evidence or search logs may remain as supporting evidence where this skill already requires them, but synthesized deliverables stay in the working packet. Update the review HTML alignment page so it renders the complete working-packet substance as structured HTML review UI: purpose-built sections, tables, matrices, gates, cards, and tier-appropriate charts or diagrams that preserve every packet section, finding, caveat, and decision detail without summary loss. Raw Markdown packet text may appear only as a supplemental source view after the rendered review UI; do not make a Full Preliminary Packet or Full Working Packet raw Markdown dump, giant <pre><code> block, link-only view, or source-only view the primary review surface. Include the evidence matrix, assumptions/confidence register, source coverage gaps, proposed canonical file changes, and artifact approval gates. Stop for either feedback-only YAML or final compiled YAML. Feedback-only YAML revises the working packet and page, then remains in Stage 2.
3. Stage 3 - Finalize approved artifacts. Consume final compiled YAML for artifact approval only when it has no unresolved needs-clarification, unresolved down feedback, or other unresolved negative feedback. Apply approved edits first, archive the working packet to docs/history/archive/YYYY-MM-DD/HHMMSS/<original-working-path>, remove the active working packet, write the approved canonical artifacts to the unchanged output paths below, and convert the alignment page to confirmed with the approval record preserved. As part of confirming, reconcile each displayed gate decision against the final compiled YAML and the written canonical artifact, render any other/freeform choice as the read-only decision and drop superseded options, and run the post-confirmation self-check from the alignment-page contract before handoff.

Canonical output paths remain unchanged. Search logs and other supporting evidence remain allowed only where this skill's output contract already requires them.

How this maps onto the Research Session Loop. This orchestrator runs as a self-advancing loop (see Execution Model below and docs/research-session-loop-convention.md): each approval stop above is a session boundary — the agent stops, the user reviews and compiles YAML, then starts a fresh Codex session and re-invokes $customer-discovery to continue. The multi-select framework page is the single scope-and-candidate approval gate (Stage 1) for the whole selected set; synthesis has its own review page. When the loop runs a framework subskill inline (state C), that multi-select approval already satisfies the framework's Stage-1 scope gate, so the framework enters at its research stage (Stage 2) and produces exactly one findings review page — one approval gate per framework. Standalone (non-loop) invocation of a framework subskill still runs its full two-stage workflow.

Evidence And Feedback Handling

Treat user feedback as input to evaluate, not as automatic ground truth.

• For factual, evidentiary, technical, or source-backed claims: verify against available evidence. If the user appears to misunderstand the evidence or states something factually incorrect, push back clearly and cite the evidence. Do not rewrite findings merely to agree.
• For taste, brand, positioning preference, risk appetite, prioritization, or other subjective judgment calls: weigh user feedback heavily and adapt the recommendation unless it conflicts with verified evidence.
• When feedback mixes facts and preference, separate them explicitly: correct the factual part, then incorporate the preference where it is a legitimate judgment call.
• When uncertain, say what is known, what is inferred, and what would change the conclusion.

Prerequisites

• Hard: None — this is the first research skill in the AFPS chain.
• Soft: Read these if they exist:
  • research/idea-brief.md (or research/{slug}/idea-brief.md) — concept, hypotheses, ICP readiness notes
  • Specs, README/CLAUDE, and relevant source files for product context
  • research/customer-feedback.md — real customer evidence (triggers product-exists mode)

Execution Model — Research Session Loop

This is a self-advancing Pattern A research orchestrator (see docs/research-session-loop-convention.md). Each invocation starts cold, resolves its state from pasted YAML + filesystem, runs exactly one heavy phase, emits the next gate, and stops. The user advances the loop by starting a fresh Codex session and re-invoking $customer-discovery. The user never invokes a framework subskill directly — the orchestrator follows each selected framework's subskill inline.

When a framework is pending, the only user-facing continuation route is re-invoking $customer-discovery with the same product/research path argument when present, for example $customer-discovery research/afps-tracker. Never tell the user to run a path-shaped child framework command; the parent resolves the pending framework from the run manifest and filesystem.

State lives in two places only:

• Run manifest — research/_working/customer-discovery-run.yaml (flat) or research/{slug}/_working/customer-discovery-run.yaml (product-path). Records the selected framework set and each framework's intermediate path. Written when the multi-select YAML is approved. Shape:

  orchestrator: customer-discovery
  slug: skills-showcase            # omit in flat mode
  selected_frameworks:
    - slug: w3-hypothesis
      intermediate: research/skills-showcase/customer-discovery-w3-hypothesis.md
    - slug: jtbd-needs
      intermediate: research/skills-showcase/customer-discovery-jtbd-needs.md
  

• Canonical-intermediate existence — a selected framework is done when research/customer-discovery-{framework}.md (or research/{slug}/customer-discovery-{framework}.md) exists, pending otherwise. pending = selected − existing-intermediates. The manifest stores selection only, not per-framework status.

research/.progress.yaml stays coarse — its pipeline_stage is a pointer, not per-framework status.

State resolution (resolve the first match; YAML first, then most-progressed A→F)

On each invocation, after Product-Path Scope Resolution (step 0), resolve state:

State	Detected when	Heavy phase this session	Emits / stops with
0 — pasted YAML	a compiled alignment YAML is pasted	branch on approval_status: ready-for-agent-review → apply the approval for the gate it answers (light: write manifest and/or prior framework intermediate, archive consumed source), then fall through to the next pending state below; not-approved → amend the named page (refinement session) and stop	amended page, or proceeds ↓
A — done	canonical research/icp.md (or research/{slug}/icp.md) exists	—	done; emit next-skill route (step 9)
B — synthesize	run manifest exists, all selected intermediates exist, no canonical icp.md (also forced by --synthesize)	synthesis (step 6)	synthesis review page
C — run framework	run manifest exists, ≥1 selected framework pending	run next pending framework inline at its research stage (step 5b)	that framework's findings review page
E — build selection	preliminary interview handoff exists, no run manifest, no multi-select page	mode detect → candidate bootstrap → build multi-select page (steps 1–5a)	multi-select review page
F — interview	nothing yet (no handoff, no manifest)	deep interview (context_intake: deep)	preliminary interview handoff, stop

State F handoff. When the deep interview completes, write the preliminary interview handoff to research/_working/preliminary-customer-discovery-interview.md (or research/{slug}/_working/preliminary-customer-discovery-interview.md) and stop — do not bootstrap candidates or build the multi-select page in the interview session. The handoff is a complete context transfer for the next cold session (state E): provisional mode signal (pre-product vs product-exists, with evidence), context summary, recommended framework subset with rationale, and all user answers that shape candidate generation and framework selection.

Light vs heavy. Recording the approved selection into the run manifest (state 0→C head), writing an already-reviewed framework intermediate, and archiving a consumed source are light — they fold into the head of the next heavy session and do not get their own round-trip. The heavy phase (interview, one framework's research, synthesis) is the only thing isolated per session.

Shortcuts. $customer-discovery discovery and $customer-discovery validate short-circuit states E→C: they write a fixed framework set straight into the run manifest after the user approves the shortcut plan, then enter state C (steps 7–8).

---

Process

0. Product-Path Scope Resolution

Resolve research scope by product path before using code or app structure as a hint:

1. If $ARGUMENTS names a non-archived research/{slug}/ directory or a product-path ID whose scope_path points there, use that path. Treat {slug} as the product/app name, not the ICP, audience, or segment label.
2. If $ARGUMENTS names only research/_archive/{slug}/ or a manifest entry with status: archived or legacy status: abandoned, stop and warn that the path is archived; do not write or update scoped outputs there.
3. Read research/.progress.yaml when present. Normalize legacy active_path to active_paths on read and write back active_paths on manifest updates. Treat legacy abandoned as archived; exclude archived, abandoned, deferred, revisit_candidate, promoted, and any scope_path under research/_archive/ from active target selection.
4. If active product paths exist in the manifest, use those paths. If multiple active paths exist, ask which one to target unless this skill explicitly supports cross-path output.
5. If no active manifest target exists, list non-archived product directories under research/, excluding research/_archive/ and dot directories. Auto-select only when exactly one exists; ask when multiple exist.
6. If no product directories exist, use flat research/ single-product mode.
7. Detect monorepo/app/package structure only as a secondary hint.

When product path {slug} is active, read and write research under research/{slug}/, specs under specs/{slug}/.

0b. Product-path manifest: Read research/.progress.yaml when present. Normalize active_path (singular legacy) to active_paths (plural list) when reading; treat legacy abandoned as archived and exclude archived/deferred/revisit/promoted paths plus research/_archive/ scopes from active target selection.

1. Mode Detection

Runs in state E (the build-selection session, after reading the interview handoff). Detect pre-product mode (default) or product-exists mode:

Pre-product mode activates when:

• No production codebase with live users detected, AND
• No research/customer-feedback.md with post-launch customer evidence, AND
• $ARGUMENTS does not contain "product" or "validate"

Available frameworks in pre-product mode:

• w3-hypothesis (default) — Schwartzfarb WHO/WHAT/WHY hypothesis generation + disproval
• jtbd-needs (default) — Ulwick/Christensen needs-based segmentation via jobs and outcomes
• four-forces (default) — Moesta Push/Pull/Anxiety/Habit switching moment analysis
• five-rings (optional) — Revella decision psychology and buyer journey. Pre-check when idea brief indicates B2B model.
• seven-dimensions (optional) — Lincoln Murphy ICP scoring with post-sale dimensions
• pmf-engine (unavailable) — requires real user data; grayed out with explanation

Product-exists mode activates when:

• Production code or deployment exists, OR
• research/customer-feedback.md exists with real customer data, OR
• $ARGUMENTS contains "product" or "validate"

Available frameworks in product-exists mode:

• pmf-engine (default) — Vohra/Supan empirical ICP from existing user data
• seven-dimensions (default) — Lincoln Murphy structured scoring with post-sale dimensions
• four-forces (optional) — switching analysis grounded in real churn/adoption data
• five-rings (optional) — buying insight refresh with real buyer evidence
• jtbd-needs (optional) — job validation against actual usage
• w3-hypothesis (unavailable) — superseded by empirical data; grayed out with explanation

2. Load Context

• Read $ARGUMENTS: if file path, read it; if text, treat as concept; if empty, check for idea brief
• Read research/{slug}/idea-brief.md or research/idea-brief.md — treat as starting hypotheses, not settled truth
• Read CLAUDE.md, README for product context
• Read codebase (if it exists): package config, key source files, routes, data models
• Read existing research/icp.md if it exists (prior run — ask user if they want to refine or start fresh)
• Read any existing research/customer-discovery-*.md intermediate artifacts (from prior framework runs)

Provisional product-path evidence. When a referenced product path is not present in research/.progress.yaml, do not treat it as a canonical active path. Require an explicit provisional-path evidence reference before using it as source context. If no such evidence exists, ask the user whether to proceed or to run $idea-scope-brief first.

3. Marketplace Side Preflight

Before identifying ICP candidates, resolve market-structure side coverage:

• Read any Market Structure Handoff or marketplace/platform/B2B2C/multi-sided notes from the idea brief, especially ## ICP Readiness. Treat side names and value exchange as hypotheses, not validated ICPs.
• If no idea brief exists, infer likely sides from $ARGUMENTS, README, specs, codebase context, or product description when the concept suggests a marketplace/platform/B2B2C/multi-sided model.
• During broad market research, validate or refute the classification with evidence.
• Before candidate generation, write a side-coverage note in the working packet.
• Candidate generation must cover each material side or explicitly explain why a side is excluded.

4. Candidate Bootstrapping

This step runs in state E only (skipped when research/icp.md already exists and user chooses to refine).

Broad market research using web search with 8–12 diverse query strategies (direct persona, pain point, market segment, trend, competitor user, forum/community, job posting, industry report, switching trigger, adjacent market, geographic/regulatory, business model, WTP signal, named account searches).

Classify the business model into one or more of: B2B SaaS (PLG), B2B SaaS (SLG), B2C, B2C subscription, marketplace/platform, B2B2C, D2C, open-source/open-core, API/developer-first.

Identify 2–5 ICP candidates. For each, note: who they are, pain evidence, accessibility, value delivery potential, WTP signal strength.

Write candidates to research/_working/preliminary-customer-discovery-research.md.

Checkpoint — Present candidates to the user. Show ICP candidates with evidence. Ask for corrections and missing segments. Incorporate feedback before framework selection.

5a. State E — Build Framework Multi-Select Page

Build the framework multi-select review alignment page with mode explanation, ICP candidates summary, available evidence, multi-select framework section (with mode-appropriate defaults pre-checked), a loop explanation (the selected set is the scope-and-candidate approval gate; each selected framework is then run inline, one findings page per framework, and the run advances by re-invoking $customer-discovery), and the approval gate.

This multi-select approval is the Stage-1 scope approval for the whole selected set. Stop for compiled YAML. Do not write the run manifest or run any framework in this session — that is state C.

5b. State C — Run Next Pending Framework (inline)

This session consumes the approved multi-select YAML (state 0→C) or advances after a prior framework's approval. At the head of the session, do the light bookkeeping first:

1. Write/append the run manifest if it does not yet exist: research/_working/customer-discovery-run.yaml (flat) or research/{slug}/_working/customer-discovery-run.yaml (product-path), recording selected_frameworks with each framework's slug and canonical intermediate path. Include only frameworks the user selected.
2. Archive the preliminary interview handoff at this selection-commit point (move it under docs/history/archive/YYYY-MM-DD/HHMMSS/<original-working-path>). A rejected multi-select page must still rebuild from the handoff, so do not archive it earlier.
3. If a prior framework's reviewed content was just approved by the pasted YAML, write its canonical intermediate research/customer-discovery-{fw}.md (or research/{slug}/customer-discovery-{fw}.md) from the already-reviewed working packet, and archive that framework's working packet and superseded review page.

Then run the one heavy phase: determine the next pending framework (first selected framework whose canonical intermediate does not yet exist), then load and follow that framework subskill's SKILL.md inline, entering at its research stage (Stage 2) — the multi-select approval already satisfied the framework's Stage-1 scope gate, so perform the research, write its working packet, and build a single findings review page. Stop for that framework's compiled YAML.

Advance the loop by self-re-invocation. The confirmed-page handoff and the terminal message name $customer-discovery and tell the user to start a fresh Codex session and re-invoke, reporting progress as "k of N frameworks complete". Do not emit cross-skill routing here — that happens only after synthesis (step 9).

6. State B — Synthesis (auto-detected; also $customer-discovery --synthesize)

Enter synthesis when the run manifest exists, all selected framework intermediates exist, and no canonical research/icp.md yet exists. An explicit $customer-discovery --synthesize also forces this state. Read all intermediate framework outputs (research/customer-discovery-{slug}.md). At least one must exist.

Synthesis mapping — how framework outputs merge into the canonical 9+1 section format:

Canonical Section	Primary Source	Secondary Sources
Customer Profile	w3-hypothesis (WHO), pmf-engine (HXC)	seven-dimensions (Ready/Willing/Able)
User Profile(s)	w3-hypothesis (WHO detail), pmf-engine (daily users)	jtbd-needs (job performers)
Trigger Events	four-forces (Push), five-rings (Priority Initiative)	jtbd-needs (unmet outcomes)
Current State Journey	jtbd-needs (current job map)	four-forces (Habit)
Pain Map	jtbd-needs (underserved outcomes)	four-forces (Push intensity)
Current Alternatives	four-forces (Habit, current solution language)	five-rings (Decision Criteria)
Market Sizing	Aggregated across frameworks	seven-dimensions (Acquisition Efficiency)
Stated Value Drivers	w3-hypothesis (WHAT/WHY), jtbd-needs (desired outcomes)	five-rings (Success Factors), pmf-engine
Customer ↔ User Dynamics	seven-dimensions (post-sale dimensions)	five-rings (Buyer's Journey)
Discovery & Evaluation	five-rings (all 5 rings)	four-forces (Active Looking)

Cross-ICP handling: Merge per-candidate evidence across frameworks, select primary ICP using seven-dimensions scoring (or value × accessibility matrix), populate Additional ICPs and Cross-ICP Analysis sections.

Build alignment page for synthesis approval. After approval: write research/icp.md and research/icp-search-log.md, then on this canonical write archive the run manifest (customer-discovery-run.yaml) and synthesis working packet under docs/history/archive/YYYY-MM-DD/HHMMSS/<original-working-path>, update research/.progress.yaml pipeline_stage to customer-discovery, and emit the downstream next-step routing (step 9).

7. State C via Discovery Shortcut ($customer-discovery discovery)

Pre-product shortcut — short-circuits states E→C with a fixed framework set (w3-hypothesis, jtbd-needs, four-forces). Build a review alignment page for the shortcut plan and require final compiled YAML approval. After approval, write the fixed framework set straight into the run manifest (research/_working/customer-discovery-run.yaml or the product-path equivalent), then enter state C and run framework 1 inline per step 5b. The loop advances by re-invoking $customer-discovery.

8. State C via Validation Shortcut ($customer-discovery validate)

Product-exists shortcut — short-circuits states E→C with a fixed framework set (pmf-engine, seven-dimensions). Build a review alignment page for the shortcut plan and require final compiled YAML approval. After approval, write the fixed framework set straight into the run manifest, then enter state C and run framework 1 inline per step 5b. The loop advances by re-invoking $customer-discovery.

9. Next Steps (after synthesis only)

Recommend the first matching condition as Recommended, remaining as Other options:

• ALWAYS: $competitive-analysis — Research competitors and market gaps for this ICP
• IF no specs/: $competitive-analysis — Map the competitive landscape
• IF specs/ exist but no research/journey-map.md: check pack availability for customer-lifecycle, recommend $journey-map if enabled, or npx skillpacks install customer-lifecycle from the project shell if not enabled
• IF codebase exists: check pack availability for business-ops, recommend $mvp-gap if enabled, or npx skillpacks install business-ops from the project shell if not enabled
• IF research/competitive-analysis.md exists: check pack availability for product-design, recommend $brainstorm if enabled, or npx skillpacks install product-design from the project shell if not enabled

10. Downstream Impact Check

After writing, check existing downstream documents for conflicts. Classify as None/Minor/Major. For Major, recommend $reconcile-research.

Output

State E output: framework multi-select review page + working packet

The run manifest research/_working/customer-discovery-run.yaml is written at the head of the first state-C session (on multi-select approval), not in state E.

State C output: per-framework findings review page + research/customer-discovery-{framework}.md

State B output (synthesis): research/icp.md (or research/{slug}/icp.md)

Canonical 9+1 section format preserved for downstream compatibility. See claude mirror for full template.

research/icp-search-log.md

Raw research log with all queries, findings, evidence, and data gaps.

research/.progress.yaml

Product-path manifest updated when secondary ICPs create parked or promotable paths.

Task Classification

• Immediately actionable → tasks/todo.md
• Human-only external actions → tasks/manual-todo.md
• Condition-gated records → tasks/record-todo.md
• Cadence-based reviews → tasks/recurring-todo.md

Constraints

• Parent self-advances one phase per invocation and follows the next pending framework's subskill inline (entering at its research stage). It bootstraps candidates, builds the multi-select selection, runs each selected framework inline, and synthesizes. The run manifest records the selected framework set; progress is the existence of canonical intermediates. The loop advances by re-invoking $customer-discovery (fresh Codex session between sessions).
• Synthesis requires at least one framework output.
• Mode detection is evidence-based.
• Stay in problem space. No features, architecture, UI, or technical solutions.
• Evidence-based. Every claim must trace back to research evidence.
• Primary ICP must use the canonical 9 top-level ## sections for downstream compatibility.
• Section 10 captures behavioural signals only.
• WTP is evidence, not pricing strategy. Route raw signals to $monetization.
• Do not overwrite existing research/icp.md without asking.
• Minimum 8 web search queries before identifying ICP candidates.
• Present before writing.

Alignment Page

When this skill produces durable deliverables (research, specs, plans, reports, prototypes, or any document output), build a full-depth HTML alignment page following ALIGNMENT-PAGE.md in this skill's directory. Output: alignment/customer-discovery-{topic}.html.

Default Shipping Contract

Follow the shared shipping contract convention in CLAUDE.md.