The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill integrates the mcp__rube__RUBE_REMOTE_WORKBENCH tool, which is designed to execute Python code at runtime.
Evidence: The documentation provides examples of using code_to_execute with import os to process files and generate documentation blocks. This pattern allows for the generation and execution of arbitrary logic within the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes external source code and files to generate documentation, creating a vulnerability to malicious instructions hidden in data.
Ingestion points: Files and directories specified in commands like /sc:document src/auth/login.js or /sc:document components/ (documented in SKILL.md).
Boundary markers: None identified; the skill does not specify delimiters or instructions to ignore embedded prompts within the code it reads.
Capability inventory: The skill has the ability to read files, write documentation, execute Python code (RUBE_REMOTE_WORKBENCH), and send data to external APIs (SLACK_SEND_MESSAGE, NOTION_CREATE_PAGE).
Sanitization: No evidence of sanitization or safety filtering for the input source code is present.
[DATA_EXFILTRATION]: The skill provides automated pathways to send processed content to external services (Notion, Confluence, Slack, GitBook).
Evidence: The mcp__rube__RUBE_MULTI_EXECUTE_TOOL is used to publish content to external platform endpoints. While intended for documentation, this creates a channel where sensitive information (like credentials or private logic found in the analyzed code) could be automatically exfiltrated if the agent is not properly constrained.

sc-document