The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input via the $ARGUMENTS variable and interpolates this data directly into a generated shell command (vidbyte retain). A malicious user could provide input containing shell metacharacters or escaped sequences to attempt to manipulate the resulting command.
Ingestion points: User-provided situation or mistake descriptions in $ARGUMENTS (SKILL.md).
Boundary markers: The generated command is wrapped in a markdown code block, but there are no instructions for the agent to sanitize or escape the input before interpolation.
Capability inventory: The skill generates shell commands for the user to execute; it does not execute them automatically.
Sanitization: No explicit sanitization or validation of the input is performed before it is placed into the command template.
[COMMAND_EXECUTION]: The skill provides a structured bash command (vidbyte retain) for the user to copy and paste into their terminal. This creates a surface for 'Copy-Paste Command Injection' where a user might be tricked into running a command that has been altered by a prompt injection attack.
[PROMPT_INJECTION]: The skill contains negative constraints such as 'NEVER name the user's specific mistake' and 'NEVER tell the user what to do next'. While flagged by automated scanners as potential concealment, these instructions are fundamental to the skill's stated purpose of providing abstract, theoretical feedback rather than practical advice and do not appear to be malicious.

theoretical-feedback