The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation specifies installing the package @jackchen_me/open-multi-agent via npm or pnpm. This is a third-party dependency sourced from a personal namespace rather than a recognized organization.
[COMMAND_EXECUTION]: The framework includes a bash tool that allows agents to execute arbitrary shell commands. This capability is demonstrated in several examples throughout SKILL.md where agents are expected to interact with the underlying system.
[DYNAMIC_EXECUTION]: The framework supports patterns where agents generate code (e.g., writing a TypeScript file to /tmp/) and subsequently execute it using the bash tool. While this is a documented feature, it constitutes dynamic execution of agent-generated content.
[INDIRECT_PROMPT_INJECTION]: The skill architecture is vulnerable to instructions hidden within the data it processes.
Ingestion points: The system ingests external data through the description fields in task definitions and via the file_read tool, as seen in the multi-agent team examples in SKILL.md.
Boundary markers: No explicit delimiters or boundary markers (such as XML tags or specific ignore instructions) are shown in the implementation examples to prevent agents from following commands found within the data.
Capability inventory: The skill provides access to powerful tools including bash for system commands and file_write for persistent file system modifications.
Sanitization: The provided code examples lack any mention of input validation or sanitization routines for data retrieved from external sources before it is passed to the agents.

open-multi-agent-orchestration