Martin Brandhaug

Four skills cover: finding and removing tautological tests that mirror implementation, writing and syncing AGENTS.md/CLAUDE.md intent-layer files, running a 7-agent codebase cleanup sweep for slop and dead code, and stripping AI writing tropes from prose like blog posts, docs, and READMEs.

Full-stack software engineers and product developers who review test suites, maintain agent context files, prepare pull requests for review, or edit prose meant for humans. They suit anyone dealing with brittle change-detector tests, AI-generated code slop, or text that sounds machine-written.

The build workflow captures intent leaf-first via SME interview; the sync workflow reconciles affected nodes after code additions, modifications, or removals. Both finish by cutting every touched node to a hard word budget so only essential architectural context remains in the hierarchy.

It targets change-detector tests that break on any refactor without catching defects: checksum assertions, mock-theater interaction tests, echo assertions, duplicate-algorithm tests, and snapshots with no oracle. Use it when writing or reviewing tests or after a behavior-preserving refactor breaks the suite.

It strips roughly 30 AI writing tells from prose, including negative parallelism, em-dash overuse, delve-family vocabulary, punchy fragments, false suspense, and signposted conclusions. Apply it to blog posts, docs, READMEs, announcements, marketing copy, and emails that sound machine-generated.