LoopFlow
Tutorial Workshop Keywords ๐ŸŽฎ LoopFlow Lab
Keywords / remember in

remember in

Cross-run memory โ€” read past lessons on start, append the outcome on stop. Augment

Syntax

remember in "<file.md>"

What it does

Names a Markdown file the loop treats as its memory. On start it reads the file and feeds those lessons into the first plan; on stop it appends a dated entry recording how the run went. Across runs, the loop learns from its own history instead of starting cold each time.

A loop forgets everything between runs. reflect gives it memory within a run โ€” a failure feeds the next plan โ€” but that context evaporates when the run ends. remember in is the across-run counterpart: it persists the hard-won lessons (which fix worked, which dead end to avoid, which flaky test needs a retry) to a file that survives, so the next invocation begins where the last one left off. Reach for it on loops you run repeatedly โ€” a nightly audit, a recurring flaky-suite chase, a migration you resume over several sessions โ€” where the same mistakes would otherwise be rediscovered each time. Because the memory is plain Markdown, it is human-readable and human-editable: you can prune stale notes, correct a wrong lesson, or seed the file with guidance before the first run. It pairs well with an after N tries guard โ€” when a loop thrashes and stops, the memory file is exactly where the next run reads why, turning a dead end into a starting advantage instead of a repeated failure.

Example

loop "keep the flaky e2e suite green":
  goal: the end-to-end suite passes three times in a row
  remember in "loop.memory.md"
  done when "pnpm test:e2e --repeat 3" passes
  each cycle: plan, then act, then observe
  when it fails: reflect, then plan again
a loop that carries lessons between runs

Common mistakes

Related