The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill uses a placeholder <name> provided by the user to determine file paths and content for multiple configuration and source files.
Ingestion points: The package <name> parameter enters the agent context and is interpolated into file paths and source code.
Boundary markers: There are no explicit boundary markers or instructions to the agent to sanitize the package name or ignore instructions embedded within it.
Capability inventory: The skill performs file system writes (creating new directories and files), modifies existing monorepo files, and executes shell commands (bun i, bun run make).
Sanitization: No sanitization or validation of the <name> input is defined, which could lead to path traversal if the name contains characters like ../ or code injection if the name is improperly escaped when inserted into files like list-of-remotion-packages.ts.
[DYNAMIC_EXECUTION]: The skill generates a build script (bundle.ts) and configuration files (package.json, tsconfig.json) based on templates and user input, and then instructs the agent to execute these generated scripts using Bun.
[COMMAND_EXECUTION]: The skill requires the execution of shell commands including bun i (installing dependencies) and a build command tsc && bun --env-file=../.env.bundle bundle.ts. The latter command specifically references an environment file (.env.bundle), which is a standard pattern for monorepo secrets management but involves sensitive file access.

add-new-package