gws-modelarmor-sanitize-response
Google Model Armor: Sanitize a model response through a Model Armor template.
Install with Tessl CLI
npx tessl i github:googleworkspace/cli --skill gws-modelarmor-sanitize-response70
Does it follow best practices?
If you maintain this skill, you can automatically optimize it using the tessl CLI to improve its score:
npx tessl skill review --optimize ./path/to/skillValidation for skill structure
Evaluation results
100%
35%AI Content Safety Wrapper
Outbound safety pipeline with correct command and template format
Correct subcommand
0%
100%
Template full resource name
100%
100%
Correct project in template
100%
100%
Correct location in template
100%
100%
Correct template name
100%
100%
Text passed via --text or pipe
0%
100%
Does NOT use +sanitize-prompt
100%
100%
README mentions outbound context
100%
100%
Without context: $0.2807 · 1m 12s · 16 turns · 71 in / 4,615 out tokens
With context: $0.3373 · 1m 5s · 22 turns · 288 in / 3,336 out tokens
100%
66%Advanced Model Response Sanitization with Structured Requests
JSON body flag usage and mutual exclusivity with --text
Uses --json flag
0%
100%
Does not combine --json and --text
100%
100%
Correct subcommand used
0%
100%
Template full resource name
0%
100%
Correct project in template
0%
100%
Correct location in template
0%
100%
Correct template name
60%
100%
comparison.md mentions --json overrides --text
33%
100%
comparison.md distinguishes the two input methods
60%
100%
Without context: $0.3218 · 1m 18s · 18 turns · 22 in / 4,671 out tokens
With context: $0.3966 · 1m 26s · 25 turns · 189 in / 4,431 out tokens
100%
60%Bidirectional AI Safety Integration Runbook
Inbound vs outbound safety command distinction and piping
Uses +sanitize-response for model output
0%
100%
Uses +sanitize-prompt for user input
0%
100%
Does NOT swap the two commands
100%
100%
Pipeline / pipe pattern shown
66%
100%
Template full resource name format
0%
100%
Correct project in template
0%
100%
Correct location in template
0%
100%
Correct template name
0%
100%
Outbound described as model-to-user
100%
100%
Inbound described as user-to-model
100%
100%
Without context: $0.6129 · 2m 38s · 39 turns · 40 in / 7,388 out tokens
With context: $0.1979 · 52s · 13 turns · 15 in / 2,685 out tokens
- Evaluated
- Agent
- Claude Code
Table of Contents
Is this your skill?
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.