The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the creation of cloud environments that can execute arbitrary code during initialization via the initScript parameter and allows for the installation of arbitrary third-party software using system-level package managers (apt, pip, npm, cargo, gem, go).
[INDIRECT_PROMPT_INJECTION]: The skill design creates an indirect injection surface where untrusted data could influence high-privilege operations.
Ingestion points: The mcp__plugin_ruflo-core_ruflo__managed_agent_prompt tool accepts a message string which is passed directly to the cloud agent runtime (SKILL.md).
Boundary markers: No explicit instruction delimiters or boundary markers are defined to prevent the managed agent from confusing data with instructions.
Capability inventory: The managed sessions have access to a persistent filesystem, network operations, and a Bash shell (SKILL.md).
Sanitization: No sanitization or filtering logic is documented for the input strings sent to the cloud environment.
[CREDENTIALS_UNSAFE]: The skill handles sensitive authentication data by allowing the specification of an authorization_token for remote MCP server connections. It also requires the management of ANTHROPIC_API_KEY or CLAUDE_API_KEY in the execution environment.
[DYNAMIC_EXECUTION]: The managed_agent_create tool enables dynamic environment configuration through the runtime assembly of package lists and initialization scripts, which are executed upon container provisioning.
[COMMAND_EXECUTION]: The skill explicitly grants access to the Bash shell within the cloud container to execute arbitrary system commands.

managed-agent