The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates commands that use the ! notation to execute shell commands. However, the analysis shows these are used as intended for gathering project context (e.g., git status, find, tree) and the skill explicitly includes validation logic (validator.py) to prevent the use of dangerous wildcard Bash permissions, requiring specific command whitelisting instead.- [DATA_EXPOSURE]: The skill interacts with local project files to generate commands and provides presets for auditing dependencies (deps-audit) or analyzing codebase structure (codebase-analyze). These operations are local to the user's environment and do not involve transmitting data to external servers.- [PROMPT_INJECTION]: The skill body contains instructional language such as 'CRITICAL' and 'NEVER ALLOWED', but these are used as part of its primary purpose to enforce safe coding standards for generated commands rather than attempting to bypass the agent's safety filters.- [DYNAMIC_CONTEXT_INJECTION]: The skill documentation and presets utilize the ! command syntax. This is the legitimate mechanism in the target platform (Claude Code) for inserting command output into a prompt. The skill uses this for benign development tasks like git log and ls.

slash-command-factory