Continue an AI task with a checked handoff, not an invented memory

In brief

A handoff should describe the state of the work, not retell the entire conversation. Distinguish completed changes from proposals and record tests as run, failed or not run.

That record makes the next assistant’s first check concrete and prevents a summary from turning an unfinished step into a false success.

Record decisions as decisions

State what the user approved, which alternatives were rejected and which question remains open. Do not write a model’s suggestion as though it were implemented. Include a concrete next action that can be checked against the current files.

Use a compact handoff

LA24X’s brief builder can organize a text draft locally; verify the filled facts yourself. Include exact test outcomes and relevant version information without credentials or unnecessary personal material.

Goal:
Constraints and confirmed decisions:
Current files/version:
Changes completed:
Checks actually run and results:
Not run / blocked:
Next concrete action:
Evidence to inspect:

Validate after the transition

Have the next assistant inspect the files and compare the stated outcome before editing. If a test result has no evidence, treat it as unverified. Update the record when scope changes rather than carrying contradictory versions of the plan forever.

What to do next

1. Record the current goal, constraints and user-confirmed decisions.
2. List the actual changed files and checks with their real outcomes.
3. Ask the next session to inspect that state before continuing, then verify its first action matches the record.