Add dual-runtime Codex and Claude architecture scaffold

.codex/README.md

@@ -0,0 +1,14 @@
+# Codex Adapter
+
+This directory contains Codex-specific workflow instructions for AI-Product-Team.
+
+Codex should use the shared pipeline definition in `../shared/pipeline.json` and produce the same session artifacts as the Claude Code adapter.
+
+## Skills
+
+- `skills/working-backwards/SKILL.md` - start or resume a session
+- `skills/wb-status/SKILL.md` - inspect session status without changing files
+
+## Status
+
+This adapter is currently scaffolded. The next implementation step is to move shared methodology, rubrics, templates, and role definitions into `shared/`, then update both Claude and Codex adapters to reference them.

.codex/skills/wb-status/SKILL.md

@@ -0,0 +1,28 @@
+---
+name: wb-status
+description: Display the current state of a Working Backwards session in Codex. Read-only.
+argument-hint: "[session-id]"
+---
+
+# Codex Working Backwards Status
+
+This skill is read-only. It must not modify session files.
+
+## Steps
+
+1. Resolve the session ID from arguments. If omitted, inspect `working-backwards/` and choose automatically only when exactly one session exists.
+2. Read `working-backwards/{session-id}/session.json`.
+3. Display:
+   - session ID
+   - feature idea
+   - created and updated timestamps
+   - current stage
+   - each stage status, critic verdict, and revision count
+   - present and missing artifacts
+4. Suggest the next command:
+   - resume if in progress
+   - review package if complete
+
+## Compatibility
+
+Use the stage list from `shared/pipeline.json` so Codex and Claude status output do not drift.

.codex/skills/working-backwards/SKILL.md

@@ -0,0 +1,52 @@
+---
+name: working-backwards
+description: Start or resume a Working Backwards session in Codex using the shared AI-Product-Team pipeline.
+argument-hint: "[feature idea] [--repo org/repo] | resume [session-id]"
+---
+
+# Codex Working Backwards Orchestrator
+
+Use this skill to run the AI-Product-Team Working Backwards pipeline in Codex.
+
+## Source of Truth
+
+Before acting, read:
+
+1. `shared/pipeline.json`
+2. `shared/README.md`
+3. Existing session state, if resuming
+
+The `.claude/` directory is useful reference material, but Codex behavior should be grounded in `shared/` so both runtimes remain compatible.
+
+## New Session
+
+1. Parse the feature idea and optional `--repo org/repo` flag.
+2. Create a session ID: `wb-YYYYMMDD-HHMMSS`.
+3. Create `working-backwards/{session-id}/session.json` using the shared pipeline stages.
+4. Start at `press-release`.
+5. Save artifacts locally by default. If GitHub persistence is requested, commit and push after each Critic PASS.
+
+## Resume Session
+
+1. Read `working-backwards/{session-id}/session.json`.
+2. Resume from `current_stage`.
+3. Preserve artifacts already marked complete.
+4. Do not restart earlier stages unless the user explicitly asks.
+
+## Stage Execution
+
+For each stage:
+
+1. Read all prerequisite artifacts.
+2. Produce or revise only the current stage artifact.
+3. Run the Critic against the stage rubric.
+4. If PASS, update `session.json` and advance.
+5. If NEEDS REVISION, revise only failing dimensions.
+6. After 3 failed revision cycles, save the best draft and pause for more evidence.
+
+## Codex Notes
+
+- Use file edits directly for generated artifacts.
+- For Visual Demo and Site Builder stages, prefer runnable React/Vite/Express output consistent with the existing Claude adapter.
+- Run local verification when available.
+- Keep Claude compatibility: session schema and artifact paths must remain the same.

AGENTS.md

@@ -0,0 +1,46 @@
+# Codex Guide for AI-Product-Team
+
+This repo defines a Working Backwards product pipeline that can run in Claude Code and Codex.
+
+Codex should treat `shared/` as the product source of truth and `.codex/` as the Codex adapter. The existing `.claude/` directory remains the Claude Code adapter.
+
+## Working Directory
+
+When working locally, use:
+
+```text
+code/AI-Product-Team
+```
+
+## Core Rule
+
+Do not skip stages. The pipeline order is:
+
+1. Press Release
+2. External FAQ
+3. Internal FAQ
+4. Visual Demo
+5. Documentation
+6. Telemetry
+7. Requirements
+
+A stage advances only after Critic PASS, subject to the max-revision pause behavior documented in `shared/pipeline.json` and the runtime adapter instructions.
+
+## Codex Responsibilities
+
+When running the pipeline in Codex:
+
+- Read `shared/pipeline.json` first.
+- Use the existing session layout under `working-backwards/{session-id}/`.
+- Preserve compatibility with Claude-generated sessions.
+- Write artifacts to disk before reporting completion.
+- For demo and site work, run available local checks when shell access and dependencies permit.
+- Do not silently resolve `[OPEN]` or `[BLOCKER]` items.
+
+## Adapter Boundary
+
+Codex-specific workflow instructions belong under `.codex/`.
+
+Product behavior, stage definitions, rubrics, templates, and methodology belong under `shared/`.
+
+If the same behavior is described in both places, prefer `shared/` unless the difference is explicitly about Codex tooling.

CLAUDE.md

@@ -1,63 +1,79 @@
 # AI-Product-Team
 
-This repository is a Claude Code extension that guides product managers through Amazon's Working Backwards methodology using a multi-agent pipeline.
+This repository is a Working Backwards product pipeline that can run through runtime-specific adapters.
+
+- `shared/` contains the runtime-neutral product definition: pipeline, methodology, rubrics, templates, and compatibility contract.
+- `.claude/` contains the Claude Code adapter.
+- `.codex/` contains the Codex adapter scaffold.
+
+The current Claude Code extension guides product managers through Amazon's Working Backwards methodology using a multi-agent pipeline.
 
 ## What this does
 
 Running `/working-backwards` starts a strict, stage-gated pipeline:
 
-1. **Stage 1 — Press Release**: Write a customer-centric Press Release before any requirements
-2. **Stage 2 — External FAQ**: Stress-test the PR with hard customer questions
-3. **Stage 2 — Internal FAQ**: Stress-test with hard engineering/leadership questions
-4. **Stage 3 — Visual Demo**: Build a working React + Express prototype of the core user journey
-5. **Stage 4 — Documentation**: Write user-facing documentation as if the product ships today
-6. **Stage 5 — Telemetry**: Define measurement and instrumentation specs
-7. **Stage 6 — Requirements**: Translate the validated artifacts into engineer-ready specs
+1. **Stage 1 - Press Release**: Write a customer-centric Press Release before any requirements
+2. **Stage 2 - External FAQ**: Stress-test the PR with hard customer questions
+3. **Stage 2 - Internal FAQ**: Stress-test with hard engineering/leadership questions
+4. **Stage 3 - Visual Demo**: Build a working React + Express prototype of the core user journey
+5. **Stage 4 - Documentation**: Write user-facing documentation as if the product ships today
+6. **Stage 5 - Telemetry**: Define measurement and instrumentation specs
+7. **Stage 6 - Requirements**: Translate the validated artifacts into engineer-ready specs
 
 Each stage must pass a Critic review before the next stage unlocks. No skipping.
 
-After each Critic PASS for Stages 1–4, the validated content is automatically published to a marketing-ready website (`site/`) that builds progressively as the pipeline advances.
+After each Critic PASS for Stages 1-4, the validated content is automatically published to a marketing-ready website (`site/`) that builds progressively as the pipeline advances.
 
 All session outputs are committed to this repo under `working-backwards/{session-id}/`.
 
+## Runtime Adapters
+
+### Claude Code
+
+Claude Code uses the existing `.claude/agents/` and `.claude/skills/` files. These remain supported.
+
+### Codex
+
+Codex uses `AGENTS.md` plus `.codex/skills/`. Codex should read `shared/pipeline.json` before running or resuming a session and should preserve the same session layout used by Claude.
+
 ## Prerequisites
 
 - `gh` CLI installed and authenticated: `brew install gh && gh auth login`
 - Git configured with push access to this repo
 
 ## Available commands
 
-- `/working-backwards [feature idea]` — Start a new Working Backwards session
-- `/working-backwards resume [session-id]` — Resume an in-progress session
-- `/wb-status [session-id]` — View current session state (read-only)
+- `/working-backwards [feature idea]` - Start a new Working Backwards session
+- `/working-backwards resume [session-id]` - Resume an in-progress session
+- `/wb-status [session-id]` - View current session state (read-only)
 
 ## Session output structure
 
-```
+```text
 working-backwards/
   {session-id}/
-    press-release.md     ← Stage 1 Critic PASS
-    faq-external.md      ← Stage 2 External Critic PASS
-    faq-internal.md      ← Stage 2 Internal Critic PASS
-    demo/                ← Stage 3 Critic PASS  (npm install && npm start → localhost:3000)
-    docs/                ← Stage 4 Critic PASS
-    telemetry.md         ← Stage 5 Critic PASS
-    requirements.md      ← Stage 6 Critic PASS
-    site/                ← built at Stage 1, updated through Stage 4  (npm install && npm run dev → localhost:5173)
-    session.json         ← updated after every agent interaction
+    press-release.md     <- Stage 1 Critic PASS
+    faq-external.md      <- Stage 2 External Critic PASS
+    faq-internal.md      <- Stage 2 Internal Critic PASS
+    demo/                <- Stage 3 Critic PASS  (npm install && npm start -> localhost:3000)
+    docs/                <- Stage 4 Critic PASS
+    telemetry.md         <- Stage 5 Critic PASS
+    requirements.md      <- Stage 6 Critic PASS
+    site/                <- built at Stage 1, updated through Stage 4  (npm install && npm run dev -> localhost:5173)
+    session.json         <- updated after every agent interaction
 ```
 
 ## Agents
 
-- `press-release-writer` — Stage 1 worker
-- `faq-writer` — Stage 2 worker (External and Internal modes)
-- `demo-builder` — Stage 3 worker
-- `docs-writer` — Stage 4 worker
-- `telemetry-writer` — Stage 5 worker
-- `requirements-writer` — Stage 6 worker
-- `critic` — Reviews all stage outputs against versioned rubrics
-- `site-builder` — Publishes validated artifacts to a marketing-ready website after each Stage 1–4 Critic PASS
+- `press-release-writer` - Stage 1 worker
+- `faq-writer` - Stage 2 worker (External and Internal modes)
+- `demo-builder` - Stage 3 worker
+- `docs-writer` - Stage 4 worker
+- `telemetry-writer` - Stage 5 worker
+- `requirements-writer` - Stage 6 worker
+- `critic` - Reviews all stage outputs against versioned rubrics
+- `site-builder` - Publishes validated artifacts to a marketing-ready website after each Stage 1-4 Critic PASS
 
 ## Rubrics
 
-Stage-specific Critic rubrics live in `.claude/rubrics/`. They are versioned JSON files — update them without redeploying any agent.
+Runtime-neutral Critic rubrics live in `shared/rubrics/`. The current Claude adapter also contains `.claude/rubrics/` for compatibility. During migration, update `shared/rubrics/` first and keep adapter copies synchronized until adapters read shared files directly.

ROADMAP.md

@@ -0,0 +1,85 @@
+# Roadmap
+
+This roadmap focuses on making AI-Product-Team a dual-runtime Working Backwards product for Claude Code and Codex.
+
+## Now: Dual-Runtime Foundation
+
+### 1. Shared Core
+
+Create a runtime-neutral source of truth for the Working Backwards pipeline.
+
+Deliverables:
+
+- `shared/pipeline.json`
+- `shared/README.md`
+- Shared locations for methodology, rubrics, templates, and agent roles
+- Compatibility contract for session artifacts
+
+### 2. Codex Adapter
+
+Add Codex-native instructions that can run the same pipeline without depending on Claude Code conventions.
+
+Deliverables:
+
+- `AGENTS.md`
+- `.codex/skills/working-backwards/SKILL.md`
+- `.codex/skills/wb-status/SKILL.md`
+
+### 3. Claude Adapter Preservation
+
+Keep the existing Claude Code extension working while reducing duplication over time.
+
+Deliverables:
+
+- Existing `.claude/` behavior remains intact
+- Future refactor plan for pointing `.claude/` files at `shared/`
+
+## Next: Product Truth Refresh
+
+The README reflects the current pipeline better than the PRD in several places. Refresh the PRD so it describes the current 6-stage flow plus dual-runtime strategy.
+
+Deliverables:
+
+- Updated `prd-pm-ai-team.md`
+- Resolved/obsolete open questions marked clearly
+- New feature backlog captured in the PRD
+
+## Next: Session Index
+
+Add a local session index so PMs can inspect all sessions without remembering IDs.
+
+Deliverables:
+
+- `working-backwards/index.md` generation guidance
+- Session status table
+- Links to artifacts, demo, docs, and site
+
+## Later: Evidence Locker
+
+Add a durable place for research inputs that agents and the Critic can cite.
+
+Deliverables:
+
+- `evidence.md` or `evidence/` folder per session
+- Evidence citation conventions
+- Critic guidance for unsupported claims
+
+## Later: Per-Team Rubric Customization
+
+Allow teams to extend base rubrics without forking the methodology.
+
+Deliverables:
+
+- Team override file format
+- Rubric version logging
+- Merge rules for base and team-specific dimensions
+
+## Later: Exporters
+
+Add selected output integrations after the core pipeline is portable.
+
+Candidate targets:
+
+- Linear issues from requirements
+- Notion page export
+- Confluence-ready documentation