The Agent Skills Directory

[COMMAND_EXECUTION]: Executes grep via the Bash tool to scan the project directories (src/, tests/, e2e/, python_backend/) for specific string patterns. This is limited to read-only search operations and is standard for code-analysis tasks.
[PROMPT_INJECTION]: The skill processes content from the codebase which could contain indirect prompt injections designed to influence the agent during the extraction and reporting phase.
Ingestion points: Reads files with .ts, .tsx, and .py extensions in multiple project subdirectories via grep and Read tools.
Boundary markers: Missing; the agent extracts metadata and descriptions directly from comments following the TODO(migration) pattern without specific delimiters to isolate potential instructions from data.
Capability inventory: Includes Read, Grep, Glob, and Bash for file system interaction and command execution.
Sanitization: The skill includes an explicit safety instruction ('Never: Automatically delete') that requires human confirmation before any code changes are made, which serves as a significant mitigation against the impact of potential malicious instructions.

migration-sweep