The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute shell commands (ls -1) to check for the presence of an overlay directory defined by the $POWERSUIT_OVERLAY environment variable. Probing environment variables and executing file system commands based on instructions can be risky if paths are not strictly validated.
[DYNAMIC_EXECUTION]: The instructions require the agent to write user drafts to a scratch file and then execute an external script (~/.claude/plugins/cache/hooks/vestige-batch.js). Executing local scripts via the shell, especially those originating from an external "plugin marketplace," introduces a dependency on the integrity of that external software.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user text via the /mode-leak command to perform style auditing.
Ingestion points: User-supplied text provided in chat or via file contents.
Boundary markers: The skill lacks explicit instructions for the agent to ignore potentially malicious instructions embedded within the text being audited.
Capability inventory: The agent is granted the ability to write to files and execute shell commands (specifically the vestige-batch.js tool).
Sanitization: There are no specific instructions to sanitize or escape user input before it is written to the scratch file used by the external execution script.

mode-leak