A ready-to-run Claude Agent SDK agent that starts every task already knowing your company. It loads your professional brain (brain/context.md, recent decisions), discovers your installed pm-skills, treats each skill's Quality Checks as its acceptance tests, and files artifacts + predictions back into the workspace.
# 1. Set up the workspace (once) — brain, context, folders:
npx pm-claude-skills init && npx pm-claude-skills add --agent claude
# 2. Copy this template out and install:
cp -R templates/apprentice ~/my-apprentice && cd ~/my-apprentice && npm install
# 3. Put it to work (your ANTHROPIC_API_KEY, your costs):
export ANTHROPIC_API_KEY=sk-ant-…
node index.mjs "draft the Q3 stakeholder update from ./notes/q3.md — CFO audience"What it does differently from a bare agent loop:
| Bare agent | The Apprentice |
|---|---|
| Starts from zero every task | Starts grounded in brain/context.md + your last 5 decisions |
| Invents document structure | Reads the matching SKILL.md first — the skill's structure is the structure |
| "Looks done" = done | The skill's Quality Checks are the acceptance test; Anti-Patterns are the reviewer |
| Output disappears into chat | Artifacts land in ./artifacts/; facts and predictions land in brain/ |
Runs on your key with acceptEdits permissions in the working directory. Read index.mjs — it's ~80 lines and meant to be forked.