lanterna-profiler

Fail

Audited by Gen Agent Trust Hub on Oct 5, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute code at runtime using the npx -y @lanterna-profiler/cli command pattern. This involves executing a package from the NPM registry that is not from a trusted organization or known vendor.\n- [EXTERNAL_DOWNLOADS]: The skill requires fetching the @lanterna-profiler/cli package from the NPM registry as part of its initialization and execution logic.\n- [COMMAND_EXECUTION]: The skill makes extensive use of shell commands to capture and process profiling data, including run, attach, and ps commands which incorporate user-supplied parameters and application commands.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests performance reports (report.agent.md, report.json) that aggregate potentially attacker-controlled data from the application under test, such as function names and file paths, and uses this data to drive investigations and build further shell commands.\n
  • Ingestion points: SKILL.md (Workflow) reads report.agent.md and report.json.\n
  • Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands within the ingested reports.\n
  • Capability inventory: The skill possesses capabilities for shell command execution ($LANTERNA run, attach) and file system writes.\n
  • Sanitization: The diagnostic workflow lacks steps to sanitize or validate the content of the profiling reports before processing.\n- [DATA_EXFILTRATION]: The profiling process involves capturing and storing sensitive metadata including environment details, working directories, and internal function names into local files (report.json, report.agent.md) which are then processed by the agent.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 5, 2026, 03:15 PM