Lesson 04 of 4 / Context

How do I know the context pack actually works?

Run one recurring job in a fresh session using only the pack and task-specific input. Record what the AI still had to ask, guessed, or misunderstood, then repair the source—not just the answer.

Build tested context routine 10 min PickBits working session

Your move

Run one job without re-explaining

The system is only useful when it survives contact with real work.

Use this rule

Run one recurring job in a fresh session using only the pack and task-specific input. Record what the AI still had to ask, guessed, or misunderstood, then repair the source—not just the answer.

Do this now

Three moves, in order.

  1. Choose a real but reversible task with a known good result.
  2. Give the fresh session the context pack and the task input without your usual verbal backstory.
  3. Compare the output against your done-list and update the brief, source, or conflict rule that caused each miss.

Build the artifact

Tested context routine

Copy these fields into a note, document, or sheet you already use.

Job The real task and known good result
Asked Useful clarification the AI requested
Guessed Unsupported assumption it made
Missed Done-list item it failed
Repair Source, brief, or rule to change

Give AI the bounded job

Prompt from your artifact—not from vibes.

Use the attached context pack for this task. Cite the source name beside material facts. Ask when a required fact is missing or conflicting. Do not use assumed background from earlier chats.

Before you use it

The work leaves your hands only when:

  • The test started without hidden verbal context.
  • Every miss maps to a repairable source or rule.
  • The finished output passes the same done-list a human uses.