hyper-google-tagmanager
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, including website HTML, GTM container exports (JSON), and API responses. This creates a surface for indirect prompt injection where malicious instructions could be embedded in the data being audited.
- Ingestion points:
scripts/gtm-detect.sh(repository files),scripts/gtm-site-check.sh(served HTML),scripts/gtm-container-audit.sh(container JSON), and Tag Manager API responses. - Boundary markers: The
Authoritysection inSKILL.mdandrules/safety-gates.md§5 explicitly instruct the agent to treat all external content as evidence, not instructions, and to report rather than follow any embedded commands. - Capability inventory: The skill has capabilities for file system access (read/search/edit), shell command execution (bundled scripts), and network operations (via
curlin scripts and API calls). - Sanitization: Data is processed using structured tools like
jqandgrep. The skill enforces a strict "evidence-based" verification workflow. - [COMMAND_EXECUTION]: The skill executes bundled bash scripts (
gtm-detect.sh,gtm-site-check.sh,gtm-container-audit.sh) to perform site and container audits. These scripts use standard utilities likecurl,jq, andgrep. - Evidence:
SKILL.mdandrules/gtm-workflow.mddefine the execution of these scripts as part of the audit and installation phases. - [EXTERNAL_DOWNLOADS]: The skill fetches content from external sources for analysis and interacts with official Google APIs.
- Evidence:
scripts/gtm-site-check.shusescurlto fetch website HTML and thegtm.jsloader fromgoogletagmanager.com. The skill also interacts with the Google Tag Manager API (tagmanager.googleapis.com). These are well-known services and are documented neutrally.
Audit Metadata