The Agent Skills Directory

[COMMAND_EXECUTION]: scripts/install.sh modifies system-level shell configuration files (.bashrc, .zshrc) to source a custom environment file, providing persistence for the toolkit.
[COMMAND_EXECUTION]: scripts/install.sh utilizes sudo apt-get for package management and executes eval on shell integration strings generated by the ntm utility.
[EXTERNAL_DOWNLOADS]: The installation script fetches and installs software from various external sources, including the public NPM registry (agentmail-cli) and a third-party Homebrew tap (dicklesworthstone/tap/dcg).
[CREDENTIALS_UNSAFE]: scripts/model-cache.sh and scripts/dispatch.sh access and utilize a large number of sensitive environment variables containing API keys for providers like OpenAI, Anthropic, Google, and others. model-cache.sh automatically probes external endpoints using these credentials to discover available models.
[PROMPT_INJECTION]: scripts/dispatch.sh includes a dispatcher command for an autonomous agent that uses --yolo and --trust flags, which are typically used to bypass safety filters and user confirmation steps.
[COMMAND_EXECUTION]: scripts/model-cache.sh attempts to execute an external binary located in another skill's directory: ~/.agents/skills/tribunal/bin/detect-runners.

agent-field-kit