@deepseek-ai/dsh-evolution-threat

Write-time threat guard for evolution tools

Model Experience

Indirect model surface

What the model sees

@deepseek-ai/dsh-evolution-threat registers no direct prompt or tool schema itself. Model-visible effects are owned by the packages that consume this service.

Token effect

Zero direct token effect from this package; consumers add any model-visible tokens.

KV Cache effect

Independent of request-prefix construction. This package does not alter the assembled prompt or tool list.

Config

• maxScanChars (default 65536): the scan window size in normalized characters. Legal range is >= 4097 — a smaller window cannot guarantee full-coverage scanning (the overlap floor is PATTERN_OVERLAP + 1 in evolution-core), so the schema rejects it at load and the assembly clamp falls back to the default. The whole text is always scanned in overlapping windows; this value is not a total cap.

• threatExemptLabels (default []): pattern labels the deployment knows to be benign. P2-4 (v14) closed the store-vs-guard asymmetry for the STORE gates (SkillLibrary/MemoryStore options) — but the exemption surface is per config site: this guard row, the tool-skill-manage row, the evolution-commands row (its own SkillLibrary for the command write paths), and the memory store options each carry their OWN list and must be set separately. A label exempted only on the store side still blocks here. (v17 audit: the evolution-commands site was missing from this enumeration.)

Known Limitations and Deferred Work

• No known durable consumer gaps at this time. Runtime contracts are covered by package and boundary tests.