Misalignment is mostly a skill issue.

This is a free library of prompt and spec patterns for people who build with AI agents. When an agent does the wrong thing, the fix is usually a clearer spec and a verification gate, not a different model. The harness you put around the model matters more than the model itself.

Every pattern has the problem it solves, how it works, a copy-paste template, and when not to use it.

Patterns

12 patterns

• Specs

  Spec-first task brief

  Write the goal, inputs, outputs and done-criteria down before the agent touches anything.

• Specs

  Acceptance criteria checklist

  Turn "done" into a list of observable, yes-or-no checks the agent and you can both verify.

• Verification gates

  Tests as the gate

  Define the check first, then let the agent iterate until the check passes. Not until it feels done.

• Verification gates

  Evidence-of-done report

  Require the agent to end with what it ran, what it saw, and what it did not verify.

• Verification gates

  Fresh-eyes review pass

  Review the diff in a new session that sees only the spec and the change, not the reasoning that produced it.

• Agent harness

  Plan, checkpoint, then execute

  Have the agent propose a plan and stop. Approve or correct it before any changes are made.

• Agent harness

  Approval gate for irreversible actions

  Let the agent draft anything it likes, but require a human yes before it sends, deletes, pays or publishes.

• Agent harness

  One task per run, small diffs

  Break work into small, separately verifiable tasks and commit between them.

• Context

  Project rules file

  Keep stable project knowledge in one versioned file the agent reads every session.

• Context

  Handoff summary between sessions

  End long sessions with a compact state file, and start the next session from it instead of the full history.

• Prompting basics

  Delimit instructions from data

  Wrap pasted documents, logs and user content in clear tags so the model can tell them apart from your instructions.

• Prompting basics

  Few-shot examples

  Show two to five worked input/output examples instead of describing the format in prose.