The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill workflow involves scanning and reading the project's source code and locale JSON files. This creates a vector where malicious instructions embedded in the codebase (e.g., in comments or UI strings) could influence the agent's behavior during the audit or replacement phases.
Ingestion points: SKILL.md prescribes the use of ripgrep (rg) on the source directory, and scripts/i18n_audit.py reads all source and JSON files from specified paths.
Boundary markers: Absent. The extraction logic (regex-based in the audit script) does not use delimiters or instructions to prevent the agent from obeying text found within the files.
Capability inventory: The skill grants the agent the ability to execute shell commands (rg), run Python audit scripts, and modify repository files (replacing strings and updating locale files).
Sanitization: Absent. No sanitization or validation of the extracted string content is performed before it is presented to the agent.
[COMMAND_EXECUTION]: The skill requires executing local shell commands (ripgrep) and a local Python script (scripts/i18n_audit.py) to perform its intended audit functions. These operations are scoped to the project's source code.

localization-toolkit