A new guide from The Rundown AI lays out a method for reviewing past Claude sessions and turning the most useful patterns into reusable skills. The process is designed to help regular users identify what is worth keeping, what should be updated, and what can be ignored.
The workflow centers on a weekly audit of recent AI activity. Instead of treating every prompt or interaction as equally valuable, the guide recommends scanning a set of recent session files for repeated tasks, recurring outputs, and other clear signs that a skill is being used often enough to justify turning it into a formal reusable asset. It also suggests leaving things unchanged when the evidence is weak.
According to the guide, the approach is meant for heavy users of Claude, Claude Code, or Codex who find themselves repeating similar prompts. It is also aimed at operators and creators who want their AI workflows to evolve over time rather than remain static. The guide says it can be useful for anyone already working with personal skills who wants a more disciplined way to decide what to build or revise.
To run the audit, users need access to Claude Cowork, Claude Code, or Codex, along with a few recent sessions or other work examples. Personal skills must already be enabled, or there needs to be a place where the assistant can save them. The guide also recommends taking a few minutes to review any proposed changes before they are created.
The core idea is to use evidence from real work instead of speculation. If a pattern appears often enough across sessions, the assistant can suggest that it be converted into a skill. If there is not enough consistency, the guide says the safest move is to do nothing.
The same general method can also be applied in Codex, according to the guide, if that is where a user keeps sessions and personal skills. That flexibility makes the workflow less dependent on a single platform and more focused on the underlying habit of reviewing actual usage.
The guide also extends the idea beyond skills and into automation. It suggests running a similar audit on scheduled tasks to identify jobs that are noisy, stale, too expensive, unclear, or missing approval steps. Here too, the recommendation is to make only changes supported by clear evidence and to accept that some setups may not need adjustment at all.
By framing audits as a regular part of AI workflow management, the guide points to a broader trend among advanced users: using the tools themselves to evaluate how well the tools are working. In this case, Claude is not just being used to complete tasks. It is also being used to study previous sessions, surface opportunities for improvement, and help create a repeatable system for refining both skills and automations.