impeccable
Fail
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The
scripts/impeccableandscripts/impeccable.cmdscripts are designed to download a platform-specific binary from an external GitHub repository (pbakaus/impeccable) and execute it. While the script includes SHA256 checksum verification, executing binary code from an unverified third-party contributor is a high-risk operation that bypasses static analysis. - [PROMPT_INJECTION]: Instructions in
reference/craft.mdexplicitly direct the agent to 'Do not tell users they need to invoke craft', which is a pattern of concealing implementation details and command usage from the user. - [COMMAND_EXECUTION]: The skill uses the
Bashtool to run local scripts and downloaded binaries. Furthermore, it usesadb exec-out screencapcommands inreference/android.mdto access and capture screenshots from connected mobile devices. - [EXTERNAL_DOWNLOADS]: The skill fetches binaries and SHA256 validation files from
github.comusingcurlandwgetoutside of the official vendor's trusted scope. - [INDIRECT_PROMPT_INJECTION]: The skill's
critiqueandauditfunctions ingest data from project files and external URLs, which creates a surface for indirect prompt injection without explicit sanitization or boundary markers mentioned in the instructions.
Recommendations
- AI detected serious security threats
Audit Metadata