hashline provides stable, verifiable line edits for LLM coding agents, replacing the usual fragile exact-string-replace approach. Instead of re-quoting old code byte-for-byte, the model reads a numbered, hash-tagged view of a file and expresses edits as line-number range operations (SWAP, DEL, INS) whose bodies contain only the new lines. Every read is tagged with a 4-hex hash of the file's content; when a patch is applied, hashline recomputes the hash and rejects the edit as "stale" if it no longer matches, so an edit never lands on content the model never saw. The core loop is read -> patch -> apply -> re-read, and the same rules work across Claude, Grok, DeepSeek, Gemini, Kimi, and any other tool-using model.