The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions in SKILL.md use a piped shell command (curl ... | sh) to download and execute a script from https://install.iii.dev/iii/main/install.sh. This method executes remote code without local verification of the script content.
[COMMAND_EXECUTION]: The references/engine-modules.md file describes the ExecModule (modules::shell::ExecModule), which enables the execution of arbitrary shell commands during the engine's startup process as defined in the iii-config.yaml file.
[EXTERNAL_DOWNLOADS]: The skill downloads an installation script from https://install.iii.dev/iii/main/install.sh and pulls Docker images from iiidev/iii:latest.
[PROMPT_INJECTION]: The skill implementation patterns for handling API requests and logs create a surface for indirect prompt injection. 1. Ingestion points: The ApiRequest structure and log trigger handlers in SKILL.md and references/api-reference.md process untrusted data from external network requests and system logs. 2. Boundary markers: There are no implementation-level boundary markers or delimiters shown in the handler examples provided in SKILL.md or references. 3. Capability inventory: The environment includes shell execution (ExecModule in references/engine-modules.md), file system writes (StateModule and KvServer in references/engine-modules.md), and network connectivity. 4. Sanitization: No input validation or sanitization routines are described or provided in the documentation or code examples.

iii