gws-modelarmor
by googleworkspacegws-modelarmor helps you work with Google Model Armor in the googleworkspace/cli ecosystem. Use it to sanitize prompts, sanitize model responses, and create templates with less guesswork than a generic prompt. It is designed for repeatable, policy-aware usage and Security Audit workflows.
This skill scores 74/100, which is acceptable for listing: it gives directory users a real, triggerable CLI entrypoint and enough workflow direction to understand when to install it, though the guidance is still fairly sparse. The repo shows a concrete `gws modelarmor <resource> <method> [flags]` interface, helper-command links, and explicit advice to inspect help/schema before execution, so an agent should have less guesswork than with a generic prompt. Users should still expect to consult adjacent shared docs and related skills for the full workflow.
- Concrete command surface: `gws modelarmor <resource> <method> [flags]` plus `gws modelarmor --help` and `gws schema modelarmor.<resource>.<method>` for discovery.
- Clear use case: filtering/sanitizing prompts and responses through Google Model Armor, with helper commands for prompt, response, and template workflows.
- Valid frontmatter and no placeholder markers, suggesting a real skill rather than a stub.
- No install command or local support files, so adoption depends on the wider `gws` CLI ecosystem and adjacent shared skill docs.
- Very limited operational detail in this file alone; users must open related skills and schema output to execute correctly.
Overview of gws-modelarmor skill
What gws-modelarmor does
The gws-modelarmor skill helps you work with Google Model Armor from the googleworkspace/cli ecosystem. Its job is to filter user-generated content for safety by routing requests through Model Armor templates and the relevant API methods.
Who should install it
Install the gws-modelarmor skill if you already use gws and need a repeatable way to sanitize prompts, sanitize model outputs, or create Model Armor templates without guessing API shapes. It is especially useful for teams doing policy-aware automation, app hardening, or gws-modelarmor for Security Audit workflows.
Why this skill is different
The main value is not a generic safety prompt. gws-modelarmor is an installable CLI skill with command discovery, schema inspection, and helper commands that reduce trial-and-error. That makes it better suited for production-style usage than a one-off prompt, especially when you need to map real inputs to required params correctly.
How to Use gws-modelarmor skill
Install context and prerequisites
gws-modelarmor install is only useful if your environment already has the gws binary available. The skill’s own instructions point to a prerequisite shared skill at ../gws-shared/SKILL.md for auth, global flags, and security rules. If that shared skill is missing in your workspace, run gws generate-skills first.
Basic invocation pattern
Use the command form shown in the skill: gws modelarmor <resource> <method> [flags]. Before you call anything, inspect the CLI surface:
gws modelarmor --help
gws schema modelarmor.<resource>.<method>
This is the fastest way to avoid malformed --params or --json input and is the core of practical gws-modelarmor usage.
Turn a rough goal into a strong prompt
A good gws-modelarmor guide input should specify three things: what content you want sanitized, which direction the output should go, and any constraints on false positives. For example, instead of “clean this text,” provide the content type, the risk you care about, and whether you are sanitizing a prompt, a response, or creating a template. That extra context helps the skill choose the right helper command and method shape.
Best files to read first
Start with SKILL.md, then read the linked shared skill in ../gws-shared/SKILL.md. After that, use gws modelarmor --help and gws schema as your live reference because this repository has no extra rules/, resources/, or helper scripts to lean on. In practice, the CLI schema is the real source of truth for current parameters.
gws-modelarmor skill FAQ
Is gws-modelarmor only for developers?
No. It is best for anyone already operating inside the gws workflow who needs safe, repeatable content filtering. Non-developers can still use it if they are comfortable reading command help and supplying structured input, but it is not a plain-language chat skill.
When should I not use it?
Do not use gws-modelarmor if you only need an informal safety review or if you are not using the gws CLI. In those cases, a generic prompt or manual review may be faster. The skill is most useful when the output must be reproducible and tied to Model Armor behavior.
How is this different from a normal prompt?
A normal prompt can describe a safety task, but gws-modelarmor gives you a CLI path, helper commands, and schema discovery. That matters when you need exact parameters, dependable invocation, or a workflow that can be reused across runs without rewriting the prompt each time.
Is this enough by itself for a security workflow?
Usually not. Treat gws-modelarmor as one controlled step in a larger review or enforcement process. For gws-modelarmor for Security Audit, you still need your organization’s policy, logging, and approval process to decide what gets flagged, blocked, or passed through.
How to Improve gws-modelarmor skill
Provide the content type and decision goal
The biggest quality gain comes from stating what you are sanitizing and what “good” looks like. For example: “sanitize customer support prompts for unsafe instructions while preserving product names” is better than “make this safe.” That reduces over-filtering and helps the model preserve useful text.
Match the command to the job
Use +sanitize-prompt for user input, +sanitize-response for model output, and +create-template when your task is policy setup rather than one-off filtering. Picking the wrong helper command is a common failure mode because the content direction matters more than the text itself.
Inspect schema before you iterate
If output is wrong, do not guess at flags. Re-run gws schema modelarmor.<resource>.<method> and compare the required fields, types, and defaults against your input. That is the fastest way to fix malformed requests and is often more useful than rewriting the prompt.
Refine with real examples
If you want better results from the gws-modelarmor skill, test it with a few representative inputs: one safe example, one borderline example, and one clearly unsafe example. Use the differences to tighten your template or prompt wording, especially if you want fewer false positives during gws-modelarmor usage.