The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the gh command-line interface to create issues on GitHub.
[DATA_EXFILTRATION]: The agent reads local codebase context and sends summarized information to GitHub, a well-known service. The instructions include safety measures to avoid including sensitive internal details like file paths or line numbers.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted user bug reports and codebase content to generate issue bodies.
Ingestion points: User chat messages and codebase files accessed during exploration (SKILL.md).
Boundary markers: No specific delimiters or instructions are used to isolate untrusted input from the prompt logic.
Capability inventory: The agent can read files and execute the gh tool to perform network operations.
Sanitization: The instructions guide the agent to interpret and rewrite the input into domain language, providing a logical filter, though no structural sanitization or escaping is implemented.

qa