The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestration script make.py executes system commands to interact with the Claude CLI and ESLint. Filenames and locale codes are strictly validated against a restrictive regex (^[a-zA-Z0-9_-]+$) before being interpolated into shell commands, effectively preventing command injection attacks. While the script uses shell=True on Windows for ESLint execution, the command arguments are derived from hardcoded constants.
[PROMPT_INJECTION]: The skill is subject to the risk of indirect prompt injection (Category 8) because it processes untrusted human-readable strings from source locale files. Maliciously crafted strings in these files could attempt to influence the agent's behavior or output format during the translation phase.
Ingestion points: English source locale files located in src/runtime/locale/*.ts are read and parsed by make.py before being passed to the AI.
Boundary markers: The script uses a structured JSON payload to delineate metadata from the translation content, which helps the model distinguish between instructions and data.
Capability inventory: The script can write to the locale directory and execute the Claude CLI and ESLint.
Sanitization: The script validates that all keys from the source dictionary are present in the output, preventing silent data loss, but it does not sanitize the natural language content of the strings for potential instructions.

bx-translate-locales