The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to use bunx oh-my-ag, which downloads and executes a package from the NPM registry at runtime. This introduces a dependency on unverified external code that could be updated or compromised independently of the skill.
[REMOTE_CODE_EXECUTION]: By invoking bunx, the skill performs a 'download-and-execute' pattern for the oh-my-ag package. This allows for arbitrary code execution on the user's system under the context of the agent's environment.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute CLI commands such as oh-my-ag agent:spawn, bunx oh-my-ag doctor, and bunx oh-my-ag dashboard. The doctor command is specifically noted for its ability to 'repair' the system, which may involve modifying environment configurations or file permissions.
[INDIRECT_PROMPT_INJECTION]: The orchestrator's design creates a significant surface for indirect injection attacks.
Ingestion points: Reads structured state and task requirements from .serena/memories/ and project files.
Boundary markers: None identified in the instructions to separate agent logic from untrusted project data.
Capability inventory: The skill has access to Read, Write, and Bash tools, can spawn new agents, and can bridge network traffic via SSE.
Sanitization: No evidence of input validation or command escaping when processing task descriptions or memory state into shell commands (e.g., inside agent:spawn).

ohmg