The Agent Skills Directory

[PROMPT_INJECTION]: The 'CRITICAL GUIDELINES' section uses imperative and authoritative language (e.g., 'MANDATORY', 'NEVER') to override the agent's default operational procedures. It specifically instructs the agent to restrict its own tool usage regarding the creation of documentation files and enforces specific path formatting rules, which constitutes an attempt to bypass or steer the agent's standard behavior logic.
[CREDENTIALS_UNSAFE]: The provided inspect_vars() function uses the declare -p command to dump all shell and environment variables. In modern development and AI agent environments, sensitive credentials such as API keys, access tokens, and secrets are standardly stored in environment variables. Providing a routine that dumps these variables increases the risk of accidental exposure of secrets to logs or the agent's execution context.
[DYNAMIC_EXECUTION]: The rollback() function template employs eval to execute strings stored in an array. This is a high-risk pattern for shell scripts, as it allows for arbitrary command execution if the content of the OPERATIONS array is influenced by untrusted input or compromised script logic during a failure event.
[COMMAND_EXECUTION]: Several provided debugging patterns, including the execute(), profile(), and breakpoint() functions, involve executing arbitrary shell commands or spawning interactive subshells. While intended for troubleshooting, these patterns significantly expand the attack surface if the agent is manipulated into executing them with malicious parameters or in unauthorized contexts.

debugging-troubleshooting-2025