# Pm Os Setup > Builds and personalizes a PM workspace that gives an agent lasting product context. It organizes vision, users, meetings, decisions, outcomes, and PRDs as indexed files; adds a Chief PM and reusable product practices; and grows with the product. Use when a PM asks for a PM OS, product workspace, product memory system, or invokes pm-os-setup. Not for application code or unrelated scaffolding. - Skill: `manansuneja/pm-os-setup` (Agent Skill, multi-file: 53 files) - Install (CLI): `npx skillmds@latest add manansuneja/pm-os-setup` - Raw SKILL.md: https://api.skillmd.com/api/skills/manansuneja/pm-os-setup/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: manansuneja (https://skillmd.com/u/manansuneja) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/manansuneja/pm-os-setup --- # PM OS ## Overview Set up a **Product Management operating system**: a plain-Markdown workspace where product context compounds instead of disappearing into chat. The PM talks to one **Chief PM** that routes work to workspace practices and compatible installed capabilities. The root `INDEX.md` is a living, human-first dashboard; meaningful content areas use local indexes so durable decisions, evidence, requirements, and outcomes remain easy for people and future agents to find. Keep the operating machinery reusable, but fit the working surface to the product's stage, the PM's scope, current priorities, users, decision cadence, and recurring outputs. Personalization models the product lifecycle and PM responsibilities, then keeps what matters, adds what is missing, and removes generated clutter. For tailored setup, read [references/product-profiles.md](references/product-profiles.md). Treat its profiles as suggestion pools, not folder bundles. Explicit requirements decide immediate scope; product and PM context provide vocabulary and high-confidence additions. ## What Gets Created ```text INDEX.md agents/pm-chief.md START_HERE.md agents/sub-agents/... AGENTS.md _workspace_setup_docs/skills/apply-pmos-struct.md CLAUDE.md _workspace_setup_docs/skills/manage-workspace-skills.md .cursor/rules/pm-os.mdc _workspace_setup_docs/personalization/... product-practices/ skills/ summarize-notes.md brainstorm.md ... templates/ meeting-summary.md outcome.md decision.md PRD templates ... references/ product-docs/ product-vision.md meetings/ outcomes/ decisions/ prds/ ``` The root and each meaningful content domain get an `INDEX.md`. Tiny leaf folders can be represented by their nearest parent index instead of carrying duplicate maps. ## Operating Modes - **Workspace mode:** When the agent has a user-selected writable workspace root, follow the full setup workflow below and create the durable Markdown workspace. - **Conversation mode:** When the client has no writable workspace root, including many mobile or web conversations, say so before setup. PM OS can still guide product work and maintain a compact Markdown `PM OS Index` in the conversation, but it must not claim that it created a persistent folder tree, a ChatGPT Project, or files that the client cannot actually access. Keep the index to current work, waiting items, recent decisions, and links or names for artifacts created in the conversation. Default artifacts to Markdown. Offer to provide supported downloadable Markdown files or a starter pack, or to continue the full setup later in a file-capable environment. Conversation mode is a useful working surface, not a silent substitute for the durable workspace. If a compatible installed capability can save an artifact to a destination the user already controls, use it only when requested or clearly appropriate and confirm the destination when it is ambiguous. Do not install, connect, or imply access to another capability automatically. ## Setup Workflow 1. **Ask one short intake.** Collect the product/project and workspace name, what is being built and for whom, current product stage and near-term outcome, the PM's active responsibilities and recurring work, existing context or preferences, and whether the current folder should become the workspace (the default) or a new folder should be created. 2. **Choose the workspace root.** Default to the current directory and rename it in place later rather than nesting a new folder inside it. A folder containing `.git/`, `README.md`, or a few starter files can still be intentional. Create `-workspace/` only when the current directory is genuinely broad or unsafe, such as a home folder, Downloads, Desktop, a repository collection, or a folder with unrelated project material. 3. **Guard against clobbering.** Never overwrite an existing workspace or unrelated project files. Treat `AGENTS.md`, `INDEX.md`, `agents/`, `_workspace_setup_docs/`, `product-practices/`, and `product-docs/` as current markers. Also recognize legacy top-level `product-skills/`. If markers exist, read `_workspace_setup_docs/workspace-state.json` when present and ask whether the user wants an update or migration instead of scaffolding over them. A newer setup skill never silently migrates an existing workspace. 4. **Run the scaffold helper.** Prefer the bundled script over writing files manually. macOS/Linux: ```bash "/scripts/scaffold-pm-os.sh" --project-name "" --target-path "." --use-current-folder "/scripts/scaffold-pm-os.sh" --project-name "" --target-path "." --create-folder ``` Windows PowerShell: ```powershell & "\scripts\scaffold-pm-os.ps1" -ProjectName "" -TargetPath "." -UseCurrentFolder & "\scripts\scaffold-pm-os.ps1" -ProjectName "" -TargetPath "." -CreateFolder ``` Do not repeatedly try unavailable shells. On macOS/Linux, call `pwsh` only when the shell helper is unavailable and `command -v pwsh` succeeds. 5. **Use a manual fallback only when scripts are unavailable.** Copy the contents of `assets/pm-os-workspace/` into the chosen root, not the template folder itself. Fill `{{PROJECT_NAME}}` and `{{DATE}}` outside `_workspace_setup_docs/personalization/`; keep tokens in that one-time guide so it remains reusable. 6. **Apply product-context tailoring.** Follow `references/product-profiles.md`. Model the users, product surfaces, lifecycle/stage, active bets, evidence, decisions, recurring artifacts, stakeholders, and things worth tracking. Turn that model into a focused mix of folders, subfolders, trackers, templates, skills, and specialists. Begin with product vision, meetings, outcomes, decisions, PRDs, and the small PM skill core—summarizing, brainstorming, synthesizing outcomes, documenting product context, and writing PRDs. Add, adapt, or remove the rest in lockstep with actual scope. Less is more. Update indexes and Chief PM routing. Keep ordinary PM artifacts Markdown-first. A request such as “create a PRD” produces `.md` by default. Use an available document, presentation, spreadsheet, design, research, or other installed capability only when the user explicitly asks for that format/capability or the context clearly requires it. If the format choice would materially change usability and intent remains unclear, ask one concise question. Do not install or connect another capability automatically. 7. **Show the result with a human-first map.** Point the user to: - `START_HERE.md` — the human welcome guide. - `product-docs/` — the primary product working surface. - `product-practices/` — skills, templates, and references controlling how work is done. - `INDEX.md` — the living workboard and high-level map. - `_workspace_setup_docs/personalization/` — agent-managed one-time setup. Explain that `agents/`, `_workspace_setup_docs/`, `AGENTS.md`, `CLAUDE.md`, and `.cursor/` are agent-facing machinery. Tell the PM they can ask the agent to add a folder, subfolder, tracker, workflow, output format, or specialist later. If they add one manually, the next agent should integrate it automatically. 8. **Offer personalization.** If the user says yes, read `_workspace_setup_docs/personalization/AGENTS.md`. Personalization must reconcile the structure, not merely add to it. It may remove clearly irrelevant generated-empty folders, skills, templates, specialists, and routes. It must ask once before deleting or relocating user-authored content. 9. **Handle root-folder renaming safely.** During personalization, rename the root folder to match the workspace name by default. Perform one same-parent rename as the final filesystem operation after all content edits and checks. Never create a destination root and move files into it. Skip the rename and ask first only if the destination exists or the root contains unrelated material. Report the old and new paths and note that the editor may need to reopen the folder. ## Conventions To Preserve - One front door: the user talks to the Chief PM; it delegates. - Keep the human surface small: `START_HERE.md`, `product-docs/`, and optionally `product-practices/`. - Mark agent-facing files clearly near the top; keep content indexes human-readable. - Keep the root `INDEX.md` as a concise workboard plus navigation. Use local indexes for meaningful content domains, substantial subdomains, and multi-file artifacts—not automatically for every tiny leaf folder. - Whenever an agent reads `AGENTS.md`, perform lightweight structural reconciliation: preserve and integrate unindexed user-created folders, subfolders, and files; add missing maps and clear routing; ask only when meaning or placement is genuinely ambiguous. - Fit the product stage and PM scope. Do not scaffold every possible PM lane. - Put synthesis, recommendations, prioritization, MVP cuts, and next steps in `product-docs/outcomes/`, not only chat. - Keep reusable instructions, formats, and examples together under `product-practices/` as `skills/`, `templates/`, and `references/`. - Determine whether guidance is task-local or a durable operating preference from its meaning and scope, not trigger phrases. When a request contains both work to produce and a reusable convention, complete the work and update the relevant existing skill/template/reference in the same task. - Keep the default library minimal. Add or update a skill for a durable preference or recurring workflow; add a specialist only when a distinct recurring role is useful. Synchronize them with content areas and Chief PM routing. - Never overwrite raw input. Preserve it under an artifact-local `raw/` folder and create clean summaries and artifacts separately. - Route specialized work through compatible installed skills or tools when useful, while keeping PM OS as the organizer. Default narrative work to Markdown; create heavier or binary formats only on explicit request or clear contextual need. Store generated deliverables with their source/context in the appropriate product area and update the nearest index. ## Notes - This skill scaffolds and optionally personalizes a PM workspace. It does not build application code. - The Chief PM is the default workspace persona. In most tools, the PM simply talks in the workspace. - Tool wiring stays minimal. `CLAUDE.md` points to `AGENTS.md`, and Cursor gets one always-on rule. Generate `.claude/agents/` or `.github/agents/*.agent.md` only when the PM asks. - `_workspace_setup_docs/workspace-state.json` records the setup version and workspace schema. Treat plugin/skill updates and workspace migrations as separate operations.