The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The managed_agent_create tool allows for the installation of arbitrary software packages from multiple registries including pip, npm, apt, cargo, gem, and go within the managed container environment.
[REMOTE_CODE_EXECUTION]: The skill enables remote code execution through the initScript parameter in the managed_agent_create tool, which runs code during the initialization of the cloud session.
[COMMAND_EXECUTION]: The skill provides access to the Bash tool and facilitates the execution of shell commands within the cloud-hosted agent environment via the managed_agent_prompt tool.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface by processing external messages that drive an agent with significant system-level capabilities.
Ingestion points: The message argument in the managed_agent_prompt tool (SKILL.md).
Boundary markers: No specific delimiters or safety instructions are defined to separate user data from agent instructions.
Capability inventory: managed_agent_create (package installation, initialization scripts) and the Bash tool (SKILL.md).
Sanitization: No input validation or sanitization logic is specified in the skill configuration.
[NO_CODE]: The skill consists entirely of markdown instructions and tool declarations in SKILL.md and does not provide any external scripts or binary files.

managed-agent