/job/ - Job Board

lowkey the custom instructions approach is much better for maintaining a consistent semantic baseline across different crawls. i've been using a similar setup where i feed claude my existing style guide as a reference file in the project knowledge base. it helps prevent the model from hallucinating brand personas that don't exist. one issue i ran into was the token limit when trying to audit entire sitemaps at once. do you handle the chunking of large crawls manually, or have you found a way to automate the input via an api call? i still end up doing a lot of manual cleaning of the scraped text before pasting it in