Method
CHAIN ACTIVE

Orient. Capture. Search. Then build.

Product Brain is not another place to paste context. The method is a working habit: load the Chain before meaningful work, capture what should survive the session, and search governed context before agents act.

Free to learn. Start with the method docs, then connect through the Quickstart. The same practice works from the CLI, MCP, and Cortex.

What changes

Most AI-assisted work starts from fragments: a prompt file, a docs folder, yesterday's chat, and whatever the human remembers to paste. Product Brain changes the starting point. The agent begins from a governed knowledge graph where decisions, tensions, standards, insights, and work packages are typed and connected.

The method is not ceremony. It is the minimum loop that keeps product knowledge current enough for agents to use during real work.

Orient

Before the agent touches the task, it orients from the Chain. Orientation answers the questions that usually live in someone's head: what is already decided, which standards apply, what tensions are still open, and what work package this task belongs to.

$ pb orient --task "Improve mobile onboarding"
Loaded: WP-42, TEN-31, DEC-89, STD-12, INS-7

Capture

Capture is where durable product knowledge leaves the chat and enters the Chain. A decision becomes a decision entry. A contradiction becomes a tension. A repeated pattern becomes a standard. The point is not to archive the conversation; it is to preserve the part future work should rely on.

CLI

Use pb capture when a terminal session produces a decision, tension, or insight.

MCP

Let Cursor, Claude Code, or Codex orient and write Chain entries without leaving the tool where work is happening.

Cortex

Review drafts, promote what is true, and inspect how entries connect across the product.

Search

Search is the guardrail before action. The agent should not only ask "what files are relevant?" It should ask "what does the Chain already know that constrains this work?" That is how a standard written weeks ago can still shape today's implementation.

$ pb search "feature flags mobile onboarding"
Found STD-12: user-visible changes require feature flags

Close the loop

At the end of a session, close it. That confirms what was captured, leaves a trail for the next person or agent, and keeps the Chain as the shared source of truth instead of another folder of notes.

Use the method where you work

Start with the docs, connect an MCP-enabled tool, or join the community to compare real workflows with other builders.