The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill provides explicit instructions for the agent to replace OpenAI library imports with a non-standard package named @claude-ai/sdk. The official SDK provided by Anthropic is @anthropic-ai/sdk. Recommending the insertion of an unverified package into a user's codebase is a supply chain attack vector that can result in the execution of malicious code during the next build or dependency installation cycle.
[COMMAND_EXECUTION]: The skill directs the agent to execute grep commands on the local filesystem to identify specific source code patterns for modification. While the commands themselves are standard, they are used here as a mechanism to facilitate the mass modification of source files to include unverified dependencies.
[INDIRECT_PROMPT_INJECTION]: The skill's migration logic creates a vulnerability surface by directing the agent to process and modify the user's codebase without implementing security boundaries or verification steps for the replacement content.
Ingestion points: The agent is instructed to read the user's source code (typically .ts files) using Grep and Read tools.
Boundary markers: Absent. The skill does not instruct the agent to verify the safety or official status of the recommended replacement package name.
Capability inventory: The agent is granted Read, Write, Edit, and Grep capabilities, enabling it to perform full search-and-replace operations on the user's filesystem.
Sanitization: Absent. The skill provides literal strings for code replacement without any validation mechanisms to prevent the injection of malicious packages.

clade-migration-deep-dive