Use when an AI agent's file reads/edits in a repo get content-blocked (refusals, "I can't help with that" on innocuous files), or before handing a repo to AI tooling, AND the repo may contain offensive-security/dual-use code (SQLi/XSS/SSRF probes, exploit/payload generators, Shodan/CVE lookups, fuzzers, credential harvesters, port scanners). Triages TRUE offensive tooling (reversibly ARCHIVE) vs FALSE POSITIVES (KEEP). Never blind-deletes.
