The destination for every product org is one AI brain with read access to every system the company runs: Slack, email, calendar, meetings, docs, source code, dashboards, CRM, design files. All of it. When an agent has access to everything, it stops feeling like a tool and starts feeling like a colleague who has been at the company for years.
Most PMs are not there yet, for three structural reasons. Procurement and security cycles take 6 to 18 months for a system that touches this many sources. The decision happens in the office of the CTO, CIO, or COO, and PMs are not the buyers. And many companies carry scar tissue from earlier knowledge-management projects. The PM agent stack is the bridge you build while that enterprise version is being procured.
What the stack actually is
It is a set of open-source Claude repos layered on top of Claude Code. The ecosystem clusters into 18 categories: awesome lists, Anthropic repos, skill collections, subagents, MCP servers, orchestration, workflows, memory, context engineering, slash commands, hooks, and more. You do not need to know every repo. You need to know which categories exist so you can find the right one when a problem hits.
Mapping the categories to the 7-stage PM OS
This is the mapping that earns its keep. Sense leans on MCP servers and scheduled-digest hooks. Discover leans on memory, synthesis skills, and researcher subagents. Decide leans on subagent collections for tradeoff analysis and consensus. Build leans on the Superpowers brainstorm-spec-plan-TDD-review loop. Ship leans on security and design-review hooks. Measure leans on postgres-mcp and playwright-mcp with scheduled variance detection. Amplify leans on writing skills and the living changelog. The mapping is rough on purpose. Memory matters everywhere, and MCP servers feed Sense, Measure, and Discover alike.
The install order
Most PMs make the same mistake: install fifteen things over a weekend and never use any of them. The order that works. Week one, the base: Claude Code, the official skills repo, github-mcp-server. Use it for a week before adding anything. Week two: Superpowers plus one MCP server for a system you actually use, and ccundo so granular undo exists. Week three: one memory tool like claude-mem, repomix, and three to five subagents that match your daily roles. Week four: walk one of the trailing how-tos (Discovery, Build, or Measure) end to end. Install five tools this week, not fifty.
What stays human, and what the stack cannot do
The stack is single-PM scope by design. It cannot read what you cannot access, memory does not propagate across teammates, and the security boundary is your laptop. You still own the strategic call, decide which problems are worth solving, and take responsibility for what ships. The agent is a force multiplier on the mechanical parts, not on judgment, taste, and accountability. Build the personal version now, and you become the person who already knows what to put on the enterprise brain when it arrives.