Series · 4 parts so far
What each release of my daily job-search routine changes, and why.
The repov1.0.0 updates only to tagged releases, stamps the engine and model on everything it writes, keeps a local git history of my workspace, and lets me steer it with commands like `why <job>` and a checklist I tick in the daily brief.
Readv0.5.0 refuses to build a cover letter with a number or a tool my own files don't back up, writes each application in a fresh helper that holds only that one job, and moves the workflows I ask for by name into skills.
Readv0.4.0 records an eval baseline on 33 labelled job ads, hands every ad to a reader that has no tools to misuse, and replaces the fit score the model picked with a rubric and a formula. German-level agreement went from 24 to 31 of 33.
ReadMy daily job-search routine runs unattended and reads job ads written by people I don't know. v0.3.0 adds permission rules and a guard hook that hold even if an ad talks the agent into something, and stops the search from losing jobs.
Read