The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to execute shell commands on a remote VPS via SSH to retrieve ranking data from an external database.
Evidence: ssh lobster@openclawtail 'export PATH="$HOME/.bun/bin:$PATH"; krankie rankings --json' in SKILL.md.
[COMMAND_EXECUTION]: The skill uses shell commands (ssh, krankie) to interact with remote services and local metadata files.
[INDIRECT_PROMPT_INJECTION]: The skill processes JSON data retrieved from a remote, non-whitelisted VPS, creating an attack surface where malicious data could influence the agent's behavior.
Ingestion points: Output from the ssh command directed to the remote host openclawtail in SKILL.md.
Boundary markers: Absent. There are no instructions provided to the agent to treat the remote command output as untrusted or to ignore embedded instructions.
Capability inventory: Shell command execution via ssh, file system read access for local metadata files.
Sanitization: Absent. The skill does not define any validation or filtering for the external JSON output before it is used to generate the ASO Audit Report.

krankie-audit