Gws Modelarmor Sanitize Response

Google Model Armor: Sanitize a model response through a Model Armor template.

Security Scan

The skill is an instruction-only wrapper around the gws CLI to run a Model Armor sanitize command and its declared requirements are coherent, but you should inspect the referenced shared auth document and the gws binary before use.

Purpose & Capability

The name/description match the runtime instructions: the SKILL.md tells the agent to run gws modelarmor +sanitize-response with a template. Requiring the gws binary is expected for this purpose; no unrelated binaries or env vars are requested.

Instruction Scope

Instructions explicitly tell the agent to read ../gws-shared/SKILL.md for auth and global flags. That delegation is reasonable for shared configuration, but it means this skill relies on another file for credentials and security rules — users should review that shared file because it may expand the skill's effective scope.

Install Mechanism

There is no install spec (instruction-only). This minimizes install-time risk — the only runtime requirement is an existing gws binary on PATH.

Credentials

This skill declares no environment variables itself (proportionate). However, the SKILL.md defers auth/global flags to ../gws-shared/SKILL.md, which may require credentials or config (not visible here). Verify that the shared file does not request unrelated secrets or broad credentials.

Persistence & Privilege

The skill is not marked always:true and uses normal autonomous invocation settings. It does not request system-wide config changes or persistent installation steps in its own instructions.

Assessment

This skill is essentially a thin wrapper that runs the gws CLI to sanitize model output — that is coherent. Before installing or enabling it: (1) Confirm you trust the gws binary on your system (verify source/version) because the skill will invoke it; (2) open and review ../gws-shared/SKILL.md as referenced — that file likely contains the auth/global flags and may request credentials or config paths (inspect any required env vars or token usage); (3) confirm the Model Armor template names you will pass are correct and that using them won’t leak sensitive outputs to unintended places; (4) if you want stricter containment, run initial tests in an isolated environment or with least-privilege credentials; (5) if you do not want the agent to call this skill autonomously, disable autonomous invocation at the agent level or avoid granting the agent access to the gws binary/credentials.

modelarmor +sanitize-response

PREREQUISITE: Read ../gws-shared/SKILL.md for auth, global flags, and security rules. If missing, run gws generate-skills to create it.

Sanitize a model response through a Model Armor template

Usage

gws modelarmor +sanitize-response --template <NAME>

Flags

Flag	Required	Default	Description
--template	✓	—	Full template resource name (projects/PROJECT/locations/LOCATION/templates/TEMPLATE)
--text	—	—	Text content to sanitize
--json	—	—	Full JSON request body (overrides --text)

Examples

gws modelarmor +sanitize-response --template projects/P/locations/L/templates/T --text 'model output'
model_cmd | gws modelarmor +sanitize-response --template ...

Tips

• Use for outbound safety (model -> user).
• For inbound safety (user -> model), use +sanitize-prompt.

See Also

• gws-shared — Global flags and auth
• gws-modelarmor — All filter user-generated content for safety commands