The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill generates specifications based on the current conversation context, which is untrusted input. It lacks explicit delimiters or instructions to disregard potential malicious commands embedded within that context.
Ingestion points: Current conversation context and codebase understanding used to produce specifications (SKILL.md).
Boundary markers: No specific markers or 'ignore' instructions are used to wrap the conversation data in the prompt templates.
Capability inventory: Shell command execution via gh and glab tools and local file writes to the .scratch/ directory (SKILL.md).
Sanitization: There are no instructions for sanitizing or escaping the data extracted from the conversation before it is used to build the specification or interpolated into shell commands.
[COMMAND_EXECUTION]: The skill utilizes external CLI tools (gh and glab) to publish created issues. It interpolates generated text directly into command-line arguments such as --title, --body, and --description. This presents a surface for command injection if the agent's shell execution environment does not adequately handle character escaping for the interpolated content from the specification.

to-spec