{
  "markdown": "# ai-pm-frame\n\nAn AI-driven Product Manager for Claude Code that transforms product materials into actionable PRDs through multi-agent stakeholder debates.\n\n## How It Works\n\nThe plugin provides one skill (`/init-pm`) that scaffolds an entire AI-PM workspace. Once initialized, two project-local skills become available inside that workspace:\n\n| Skill | Scope | What It Does |\n|-------|-------|-------------|\n| `/init-pm` | Plugin-level | Analyzes documents or source code, generates a PM persona, and scaffolds the workspace |\n| `/debate` | Project-local (installed by init) | Runs a live, turn-by-turn adversarial debate with 5+ AI stakeholder roles |\n| `/gen-prd` | Project-local (installed by init) | Synthesizes all context into a production-grade PRD |\n\n## Installation\n\n```\n/plugin marketplace add ATreep/ai-pm-frame\n/plugin install ai-pm-frame@ai-pm-frame\n```\n\nVerify with `/plugin marketplace list`.\n\n**Source code mode** requires an additional plugin:\n\n```\n/plugin marketplace add ATreep/one-workflow\n/plugin install one-workflow@one-workflow\n```\n\nIf you only provide documents, no extra plugins are needed.\n\n## Quick Start\n\n### 1. Initialize an Empty Project\n\nCreate an empty directory for your product workspace, open Claude Code there, and run:\n\n```\n/init-pm\n```\n\nTwo input modes — use whichever is available, or combine both:\n\n- **Documents:** PRD, ads, user stories, competitive analysis, product docs\n- **Source code:** Run inside an existing codebase. Claude scans source, generates a detailed `spec/` folder, and derives the PM role\n\nThis creates:\n\n```\nyour-project/\n├── pm-role.md                  # Concise AI PM persona (100-200 lines)\n├── claude-pm.sh                # Launch script\n├── .claude/\n│   ├── settings.json           # Agent team config\n│   └── skills/\n│       ├── debate/SKILL.md     # Debate skill (project-local)\n│       └── gen-prd/SKILL.md    # PRD generation skill (project-local)\n├── debate-materials/           # Drop debate inputs here\n├── prd-outputs/                # Generated PRDs land here\n└── spec/                       # (Source mode only) Architecture, modules, data model...\n```\n\n### 2. Add Debate Materials\n\nCopy research docs, competitor briefs, user interview notes, or any relevant files into `debate-materials/`.\n\n### 3. Start the AI-PM Shell\n\n```bash\n./claude-pm.sh\n```\n\n### 4. Run a Debate\n\n```\n/debate Should we build a mobile app or a PWA for the next quarter?\n```\n\n### 5. Generate a PRD\n\n```\n/gen-prd\n```\n\nThe PRD reads from `pm-role.md`, `spec/`, debate outputs, materials, and source code to produce a structured document with requirements, user stories, data models, success metrics, and more.\n\n## How the Debate Works\n\nThe debate is a real-time, sequential multi-agent process — not a summary generator. Only available inside a workspace initialized by `/init-pm`.\n\n1. **Role generation** — Claude reads your materials and generates 5+ stakeholder roles with distinct incentives, metrics, and constraints. Always includes at least one product role and one commercial role.\n2. **Sequential turns** — Each role speaks one at a time in a fixed canonical order. No parallel speaking, no batching.\n3. **Adversarial by design** — Roles challenge named opposing claims with evidence, expose contradictions, and escalate into hard trade-off analysis.\n4. **Progressive depth** — Early rounds surface positions, middle rounds expose contradictions, late rounds force decision criteria and implementation consequences.\n5. **Verbatim transcript** — Every turn is relayed to you in real time and saved to a markdown file.\n6. **Final report** — After 10+ rounds, the facilitator synthesizes a structured decision assessment with recommended next steps.\n\nOutput is saved to `debate-outputs/debate_output_<topic>_<YYMMDDhhmm>.md`.\n\n## How PRD Generation Works\n\n`/gen-prd` is a project-local skill installed by `/init-pm`. It synthesizes all available sources in priority order:\n\n| Priority | Source | What It Provides |\n|----------|--------|------------------|\n| 1 | `pm-role.md` | Product vision, target audience, core features, business goals |\n| 2 | `spec/` | Architecture, modules, data models, integrations |\n| 3 | Source code | Implemented features, API surfaces, technical constraints |\n\nThe PRD includes 17 mandatory sections: executive summary, problem statement, personas, user stories with acceptance criteria, functional and non-functional requirements, data model, API design, release strategy, success metrics, risks, and more.\n\nEvery requirement must pass the \"engineer test\": can an engineer estimate and build this without a follow-up meeting?\n\nOutput is saved to `prd-outputs/prd_<product-name>_<YYMMDDhhmm>.md`.\n\n## Project Structure\n\n```\nai-pm-frame/\n├── .claude-plugin/\n│   ├── plugin.json             # Plugin metadata\n│   └── marketplace.json        # Marketplace listing\n├── assets/\n│   ├── debate/SKILL.md         # Debate skill template (copied to project on init)\n│   └── gen-prd/SKILL.md        # PRD generation skill template (copied on init)\n├── skills/\n│   └── init-pm/SKILL.md        # Workspace initialization skill\n└── README.md\n```\n\nSkills in `assets/` are installed as **project-local** skills (`.claude/skills/<name>/SKILL.md`), not at the user level (`~/.claude/`). This keeps them version-controllable and customizable per product.\n\n## Requirements\n\n- Claude Code with agent team support (`CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1`)\n- Product context from one of: documents (PRD, ads, docs) or an existing codebase\n- **One Workflow plugin** — only required for source code mode. Not needed for documents.\n\n## License\n\nMIT\n",
  "bytes": 5608,
  "sha": "ce779183cc918c591e8a542713c0265ee5578877e1448e1b2f7be5a79ad24309",
  "repo_slug": "atreep/ai-pm-frame",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_atreep_ai_pm_frame_ai_pm_e2979855/readme"
}