Skip to content
Canon

Playbooks

Operational procedures agents can run.

Numbered steps, a done-when, a refuse-list. Not essays.

  1. Design an agentEvals first. Raw loop second. Framework last, and only for a missing primitive.12 min · 5 steps
  2. Choose a topologyWalk the decision tree. Do not start from a crew because the demo looked collaborative.6 min · 3 steps
  3. Write a skillFrontmatter that triggers, a body that runs, files on demand. Distill from traces if you have them.10 min · 3 steps
  4. Write AGENTS.mdThe constitution of the repo. Commands, boundaries, done. Point at skills for depth.8 min · 3 steps
  5. Design the ACINames, schemas, errors, caps, idempotency, the permission gate. This is the real prompt.10 min · 3 steps
  6. Add a human gateApproval and missing info are tools. Pause is an API. The blob survives the wait.8 min · 3 steps
  7. Spawn subagentsA nested loop to protect the parent window, to parallelize, or to apply a different skill set.8 min · 3 steps
  8. Own the windowBuild the working set every turn. Pin the goal. File the bulk. Compact the errors.8 min · 3 steps
  9. Write evals firstThe golden set is the product specification. Architecture waits on it.10 min · 3 steps
  10. Review a designPaste the design. Score it against named failures and the five readiness questions.5 min · 3 steps
  11. Sandbox a coding agentWorkspace, permissions, git as ledger, verifier in the loop. Do not start from a blank graph.12 min · 3 steps
  12. Defend injectionUntrusted speakers everywhere: pages, PDFs, MCP descriptions, other agents. Executor enforces.8 min · 3 steps
  13. Distill a traceCompile a paid-for lesson into a skill, a rubric, or a golden case. Do not hoard transcripts.8 min · 3 steps
  14. Pick a frameworkPick for the constraint you have. Steal modular concepts. Do not rewrite the company around a README.6 min · 3 steps
  15. Ship durableThe reducer over a blob, with pause, resume, traces, and evals. This is production.10 min · 3 steps