The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The tool's primary scripts (check-catalogs.mjs, check-i18n-code.mjs) ingest and process application source code and JSON catalog files which are likely to contain translated text from external contributors or automated systems.
Boundary markers: No explicit boundary markers are used when interpolating catalog values into linter reports or agent context, which could allow a malicious translation to attempt to override agent instructions.
Capability inventory: The skill has the capability to write to arbitrary paths within the project directory (scripts/sync-copy.mjs) and execute the formatjs CLI tool (templates/scripts/i18n/verify-catalogs.ts).
Sanitization: Content is validated for syntax and script correctness (Rule L2, Rule L10) but is not sanitized against potential instruction-based payloads directed at the AI agent.
[OBFUSCATION]:
Static analysis signals regarding hidden Unicode characters were evaluated and found to be benign. The characters in question (U+2066 through U+2069, U+061C) are bidirectional text control characters used as literal search patterns in linter rules (e.g., Rule L8: 'bidi controls/ALM in catalog').
Analysis confirmed that these characters are essential for the skill's purpose of ensuring proper UI display of bidirectional text and are not being used for steganographic instruction hiding.

i18n-strings-and-catalogs