Questlog

Agents forget. That's the central problem. You hand one a six-step task, it does step one, drifts through step two, invents step three, and by step four it's quietly pretending the first three are done. Not because the model is bad — because the model has no persistent place to write down what it's doing and check back against it.

@ghostpaw/questlog is a lightweight npm library for giving an AI agent a persistent quest log — a place to record multi-step work, mark progress, and keep the plan in sight across a session. It's the digital equivalent of the paper checklist a good contractor keeps on the clipboard: small, dumb, durable, and the reason the job finishes.

It's aimed at developers building agents that do work bigger than a single prompt. Research pipelines, multi-file refactors, research-and-write flows, anything where "the agent forgot what it was doing" is a failure mode you've already hit once. Instead of keeping the whole plan in the model's context window and praying, you keep it in questlog, and the agent reads from and writes to it as it works.

The library is small on purpose. It isn't an agent framework and it isn't a database — it's one narrow abstraction: a log of tasks with states. You wire it into whatever loop you've already got.

Pair it with a runner like Neoagent for quick one-shots that still need a breadcrumb trail, or with a multi-agent framework like Squad Station 2 when the log needs to be shared between agents.

If your agent's main failure mode is "lost the plot," add questlog and see how much of that goes away.