Mac Anderson

Field manual

Self-assessment

Answer yes or no. Count a yes only if you could show the evidence today. Each no points at the part of the field manual that fixes it. Score yourself now, and again after you have worked through the parts.

  1. 1Can you split one run's tokens into context construction and reasoning?
    If no, read part 1
  2. 2Does code, not the model, reduce a stack trace before it reaches the window?
    If no, read part 2
  3. 3Are file paths and symbols in a prompt resolved against an index before the first model call?
    If no, read part 3 and 4
  4. 4Does a new run start from stored knowledge of the repository?
    If no, read part 5
  5. 5Can every standing instruction fail a check in CI?
    If no, read part 6
  6. 6Does every context assembler take a token budget?
    If no, read part 7
  7. 7Does the agent see the live schema, with lineage, before it writes a migration?
    If no, read part 8
  8. 8Is control flow between steps written as code?
    If no, read part 9
  9. 9Can you look up the tests that cover a function?
    If no, read part 10
  10. 10Does old content leave the window by policy?
    If no, read part 11
  11. 11Can you answer a three-hop question about your system with one query?
    If no, read part 12
  12. 12Do you report cost per completed task, including retries?
    If no, read part 13
  13. 13Can you list every agent that can write to a system, with a named operator for each?
    If no, read part 14
  14. 14Is each agent's authority, budget, and equipment written in one reviewed place?
    If no, read part 15
  15. 15For a given write action, can you name the rule that allowed it and its author?
    If no, read part 16
  16. 16Can you attribute last month's spend to the agent, the run, and the person?
    If no, read part 17
  17. 17Do you know how many tools each agent is offered, and how many it calls?
    If no, read part 18
  18. 18Could someone outside your team reconstruct one agent action in 30 minutes?
    If no, read part 19
  19. 19Are completion checks for bounded tasks fixed before the run starts?
    If no, read part 20
  20. 20Does each ongoing agent have a review date on a calendar?
    If no, read part 20