MadsLorentzen/ai-job-search
This Claude Code framework is a concrete agent-harness pattern: structured source files, specialized commands, a drafter-reviewer loop, compilation checks, and explicit human approval boundaries.
The local-first framework covers profiling, job discovery, fit ranking, tailored applications, and interview preparation. Its core pipeline uses a **second reviewer agent** and a **PDF compile loop**; the author reports 69 applications, 20 first interviews, and one contract.
Study it as a reusable harness pattern: keep source data in files, split work into named skills, give agents explicit evaluation criteria, and verify generated artifacts with deterministic tools before presenting them. Human approval remains in the tracking and messaging flows.
The local-first framework covers profiling, job discovery, fit ranking, tailored applications, and interview preparation. Its core pipeline uses a **second reviewer agent** and a **PDF compile loop**; the author reports 69 applications, 20 first interviews, and one contract. Study it as a reusable harness pattern: keep source data in files, split work into named skills, give agents explicit evaluation criteria, and verify generated artifacts with deterministic tools before presenting them. Human approval remains in the tracking and messaging flows. Its portal integrations are mainly Danish, and instruction-level prompt-injection defenses are **not a sandbox**. A public fork can expose tracked personal data, so personal use should start from a private repository with this project configured as upstream.
This turns familiar harness patterns into an end-to-end, consequential personal workflow with reported funnel outcomes: file-backed state, named skills, a second reviewer, deterministic PDF checks, and human approval. It reinforces structured delegation without validating general hiring effectiveness; Danish integrations, prompt-injection exposure, and the risk of publishing personal data materially narrow reuse.