noatmark-dsh-plugin

In LLM conversations or automated pipelines, pasted text and upstream model output rarely arrive clean: zero-width characters, hidden markers, stray formatting fences, and formula-injection prefixes sneak in unnoticed. noatmark-dsh-plugin packages the deterministic engines behind noatmark.com into four tools your dsh agent can call on demand. sanitize_text strips invisible characters and flags prompt-injection signals in one pass; scan_text reports every zero-width character with its code point and position, plus any detected injection patterns or hidden text; clean_format removes common LLM formatting artifacts such as extra blank lines, orphaned code fences, and trailing whitespace without touching meaning; and sanitize_csv prepends a leading quote to values starting with =, +, -, or @, following OWASP guidance to neutralize CSV formula injection. Everything runs locally with deterministic, reproducible results, so no data ever leaves your machine. The plugin is a natural fit for security teams auditing untrusted input, data engineers building CSV ingestion pipelines, and any developer who needs a trust-and-verify step between raw LLM output and downstream processing.