- A Claude account (claude.ai)
- A terminal or IDE with a terminal
Set up the kit.
~15 min · run once before Day 1Install Claude Code
Follow the instructions at claude.md.
Clone the PM Agent Kit
git clone https://github.com/nicoladevera/pm-agent-kit
Download the Terrain company context and place it in the kit
Run the one-liner below, or download the ZIP manually. If downloading manually: unzip it, then copy the contents of the extracted folder (not the folder itself) into pm-agent-kit/company/ — this replaces the stub files already there.
curl -L https://nicoladevera.github.io/pm-agent-kit-workbook/downloads/terrain-company-context.zip -o /tmp/terrain.zip && unzip -o /tmp/terrain.zip -d pm-agent-kit/company && rm /tmp/terrain.zip
This replaces the kit's stub placeholders with pre-filled Terrain context so every skill runs at full capacity. If you have already added real company context to pm-agent-kit/company/, back it up before running this.
Open Claude Code in the pm-agent-kit directory
cd pm-agent-kit claude
Run a smoke test to confirm the kit is working
/doc-review Review this and tell me if it's a strong PRD: "We want to add a search filter. Users can't find things."
If the skill returns structured feedback, you're set.
The kit works through skills — each is a slash command like /prd-draft or /doc-review that reads your company context files and your input, then produces a structured PM artifact.
Output quality improves significantly when the agent has product, team, customer, and process context. That's why copying the Terrain files matters: it loads real context into the kit so you experience it at full capability from Day 1, not a watered-down version.
You don't need everything installed on Day 0. Install the extras before the day they're needed.
For full documentation on how the kit works, see the PM Agent Kit README on GitHub →