krankie-rankings
Warn
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on
sshto execute multi-part shell commands on a remote server (openclawtail) to perform keyword checks. - [REMOTE_CODE_EXECUTION]: The agent is instructed to execute an external script (
~/.krankie/kr_report.mjs) using thebunruntime on the remote host. The source code for this script is not provided in the skill package, making its behavior opaque. - [DATA_EXPOSURE]: Internal infrastructure details are exposed, including a specific username (
lobster), a hostname (openclawtail), and absolute directory paths on both a remote VPS and a local MacBook (/Users/antonnovoselov/Documents/Vault/...). - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the public App Store search API.
- Ingestion points: Data retrieved via
krankie rankings --jsonand written to files read by the agent. - Boundary markers: No delimiters or instructions to ignore embedded content are present.
- Capability inventory: The agent has SSH shell access, script execution via Bun, and file transfer capabilities via
scp. - Sanitization: No sanitization or validation of the ranking data is described.
Audit Metadata