mobile-launch

Warn

Audited by Gen Agent Trust Hub on Mar 23, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill enforces an autonomous operational mode ('Do NOT ask the user questions') and a 'Self-Healing' validation phase that automatically fixes code and re-runs builds. This removes human-in-the-loop verification for intrusive host-level operations.
  • [COMMAND_EXECUTION]: The skill performs automated application builds, compilations, and test suite execution. When combined with the $ARGUMENTS input, this creates a potential command injection surface on the developer machine.
  • [DATA_EXFILTRATION]: The skill accesses the platform-specific sensitive directory '~/.claude/projects/' to record execution telemetry in 'skill-telemetry.md'. This involves persistent tracking of project metadata and execution state.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection via its primary input vector. 1. Ingestion points: Untrusted data enters via the '$ARGUMENTS' variable. 2. Boundary markers: None present to isolate input from instructions. 3. Capability inventory: Code modification, shell command execution (build/test), and file system writes (~/.claude/projects/). 4. Sanitization: No validation or escaping of input content.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 23, 2026, 10:58 AM
Security Audit — agent-trust-hub — mobile-launch