native-app-profiling
Warn
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: MEDIUMPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill's 'Gotchas' section explicitly mentions that operations may need to run with
sudo. Instructions directing an agent to use administrative privileges increase the risk of unauthorized system modifications or access. - [INDIRECT_PROMPT_INJECTION]: The skill provides command templates and a bash script that interpolate external inputs (such as application names, PIDs, and durations) directly into shell execution strings. This creates a surface for command injection if the input contains shell metacharacters.
- Ingestion points: User-supplied application names, PIDs, and durations in
SKILL.md. - Boundary markers: None; variables are interpolated directly into shell templates.
- Capability inventory: Shell command execution using
xcrun,pgrep,vmmap, andatos. - Sanitization: Minimal; while some variables in the bash script are quoted, the overall pattern relies on the agent providing 'safe' strings without internal validation.
- [COMMAND_EXECUTION]: The skill utilizes
vmmapto inspect the memory layout of running processes. While required for the stated purpose of symbolication, this involves accessing sensitive process metadata including memory addresses and ASLR offsets.
Audit Metadata