hyperframes
Fail
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill directs the agent to execute
npx hyperframes@latest upgradeandnpx hyperframes skills update. These commands download and run code from a third-party registry (npm) without pinning to a specific version or cryptographic hash, allowing for potential execution of malicious code if the package is compromised. The instructions explicitly tell the agent to "Act on the signal rather than relaying it to the user," which bypasses human review for high-risk software modifications.\n- [EXTERNAL_DOWNLOADS]: The skill fetches and installs specialized workflow components and "domain skills" from theheygen-comrepository and npm registry, which are external third-party sources relative to the attributed author.\n- [COMMAND_EXECUTION]: The skill mandates the use of shell commands includingnpx,node, andgh(GitHub CLI) to manage project state, capture website content, and interact with external data sources.\n- [DYNAMIC_EXECUTION]: The skill executes local scripts using paths resolved at runtime, such asnode <MEDIA_DIR>/scripts/prefs.mjs. This dynamic resolution and execution of scripts poses a security risk if the directory variables are manipulated.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection through its ingestion of untrusted external content.\n - Ingestion points: Data enters the agent's context through GitHub Pull Requests (
gh pr view), website crawls (npx hyperframes capture <URL>), and Figma design URLs.\n - Boundary markers: The instructions lack explicit delimiters or warnings to ignore embedded instructions within the ingested PR text or website content.\n
- Capability inventory: The skill has access to shell command execution and file system writes.\n
- Sanitization: There is no evidence of sanitization or filtering for the external text content before it is processed by the agent.\n- [METADATA_POISONING]: The skill includes an internal file (
skill-report.json) that identifies the author asheygen-com, which contradicts the system's attribution toaiskillstore. This discrepancy in authorship metadata is deceptive and could lead to a misjudgment of the skill's safety and provenance.
Recommendations
- AI detected serious security threats
Audit Metadata