The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The documentation describes platform features that ingest untrusted data into the agent's context. Ingestion points include the use of positional arguments ($1, $2), dynamic arguments ($ARGUMENTS), and file references (@path/to/file) as defined in SKILL.md. Boundary markers such as the recommended use of 'argument-hint' and shell-based validation logic are discussed. Capability inventory includes access to view_file, write_to_file, and run_command tools. Sanitization is addressed through examples in references/advanced-workflows.md showing input validation using grep and shell conditionals.\n- [DYNAMIC_CONTEXT_INJECTION]: The skill documents the syntax used to incorporate shell command output at command load time. Evidence includes educational content in SKILL.md and references/plugin-features-reference.md regarding the use of backticks for shell invocation. This is a legitimate documentation of a platform-native feature for developers.\n- [COMMAND_EXECUTION]: Examples of system utility commands are present within the testing and maintenance documentation. Evidence in references/testing-strategies.md demonstrates the use of dd for generating test files and rm for clearing temporary files in the /tmp directory. These instances are strictly instructional and do not represent a persistence or destruction threat.

command-development