A Short Course

ClaudeTag

Learn Anthropic's Slack-native agent — and steal its ideas for writing better agent prompts.

8 lessons · ~10 min each · grounded in Anthropic primary sources · built for a live-systems operator
Start here New to ClaudeTag? Read the Field Guide first (~8 min, the whole picture), then work the lessons in order. Each is short, ends in one concrete win, and has auto-graded retrieval questions. Every lesson lands back on the same mission: delegation done well — scoped, budgeted, audited.

The lessons

  1. 01The Delegation ContractThe four clauses behind every setup: objective, tools, instructions, budget.
  2. 02Scoping a prism-ops ChannelBuild a real least-privilege Access bundle. Read vs mutate; the kill_global edge.
  3. 03Ambient Mode: the Guardian that Speaks FirstProactive alerts without noise. Define "normal"; the escalation ladder.
  4. 04Multiplayer & Separate IdentitiesOne Claude per channel; a fleet of scoped identities with walled memory.
  5. 05Writing the @Claude TaskA delegation prompt that decomposes cleanly. Objective, success, constraints, output.
  6. 06The Memory LifecycleCurate what it learns; keep secrets out; the privacy trade-off.
  7. 07Setup, Spend & the Audit LogOwner-only setup, fail-closed spend caps, and proving what it did.
  8. 08CapstoneClaudeTag vs Claude Code, and an interleaved self-test across the course.

Reference

ClaudeTag — A Field Guide — the compressed, fully-cited reference. Keep this one; the lessons you'll finish and move on from.

Primary sources