The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute multiple shell commands, including find, grep, and node, to inventory files and run a provided codemod script (mintlify-codemod.mjs). It also performs package management operations using pnpm install to update the project's dependency tree and lockfile.
[INDIRECT_PROMPT_INJECTION]: The skill's core function is to process content from a user-provided documentation repository, which presents a surface for indirect prompt injection if the source content contains malicious instructions designed to exploit the agent's capabilities.
Ingestion points: The agent reads the entire directory structure and file contents of the source repository, including markdown/MDX files and configuration files such as docs.json, docusaurus.config.js, and meta.json.
Boundary markers: The instructions do not specify any boundary markers or special handling to isolate the content of the source files from the agent's own instructions during processing.
Capability inventory: The skill has significant capabilities, including reading and writing arbitrary files within the project, executing local scripts, and running package manager commands that could potentially trigger arbitrary code execution via hooks.
Sanitization: The skill uses regular expressions and a Node.js script to transform data but lacks comprehensive sanitization or validation of the input content before it is processed or used in shell commands.

blume-migrate