caam
Fail
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPERSISTENCECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The installation instructions in
SKILL.mduse acurl | bashpattern to execute a script directly from a remote GitHub repository. This gives the external script full execution privileges on the user's system. - [EXTERNAL_DOWNLOADS]: The setup process downloads an installation script from
raw.githubusercontent.com/Dicklesworthstone/coding_agent_account_manager/main/install.sh. - [DATA_EXFILTRATION]: The skill is designed to read, store, and restore high-value authentication files including
~/.claude.json,~/.config/claude-code/auth.json,~/.codex/auth.json, and~/.gemini/oauth_credentials.json, which contain session tokens for paid AI services. Additionally, the 'Isolated Profiles' feature symlinks the user's~/.sshand~/.gitconfigdirectories into its own vault, exposing private keys and configuration. - [PERSISTENCE]: The documentation instructs the agent or user to modify shell profile files (
~/.bashrcor~/.zshrc) to create aliases for standard CLI tools. This ensures the skill's wrapper logic intercepts and manages sessions across all terminal sessions. - [COMMAND_EXECUTION]: The
caam runandcaam execcommands facilitate tool management by launching subprocesses with manipulated environment variables such as$HOMEand$CODEX_HOME.
Recommendations
- HIGH: Downloads and executes remote code from: https://raw.githubusercontent.com/Dicklesworthstone/coding_agent_account_manager/main/install.sh?$(date - DO NOT USE without thorough review
Audit Metadata