The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documents common installation script patterns that use curl or wget to fetch binaries from external sources.
Evidence: references/plugin-distribution.md contains a template for downloading a binary from a GitHub repository using curl -fsSL "$url" -o "${BIN_DIR}/mytool".
[REMOTE_CODE_EXECUTION]: Patterns for dynamic shell integration are provided, which involve executing code generated by external binaries.
Evidence: references/plugin-distribution.md demonstrates the eval "$(mytool init zsh)" pattern common in modern CLI utilities to set up environment hooks and completions at shell startup.
[COMMAND_EXECUTION]: Illustrative examples of Inter-Process Communication (IPC) using socat are included.
Evidence: references/posix-scripting.md includes a socat listener example (socat UNIX-LISTEN:/tmp/myapp.sock,fork EXEC:./handler.sh) used to demonstrate Unix socket communication. While flagged as a potential reverse shell pattern by automated scanners, it is a legitimate IPC pattern within the context of the documentation.
[PERSISTENCE]: The skill describes methods for modifying shell initialization files to maintain plugin functionality across sessions.
Evidence: references/plugin-distribution.md contains functions like add_to_shell_config and install_to_shell that append initialization logic to .bashrc, .zshrc, and Fish configuration directories.
[INDIRECT_PROMPT_INJECTION]: The skill documents hooks that process command-line input and shell events, creating a potential surface for indirect prompt injection if an agent interacts with untrusted inputs through these patterns.
Ingestion points: Shell events and command strings are ingested via PROMPT_COMMAND (Bash), preexec/precmd hooks (Zsh), and event handlers (Fish) as shown in references/bash-integration.md, references/zsh-integration.md, and references/fish-integration.md.
Boundary markers: The provided examples do not include explicit delimiter-based boundary markers or instructions to ignore embedded commands in the processed data.
Capability inventory: The skill demonstrates capabilities for file writing (references/plugin-distribution.md), network operations (references/plugin-distribution.md), and subprocess execution (references/posix-scripting.md).
Sanitization: The documentation focuses on functional implementation and does not demonstrate specific sanitization or escaping of external command content to prevent injection.

shell-integration