The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external source files to extract terminology, creating a potential surface for indirect prompt injection.
Ingestion points: The agent reads source strings from catalog files (e.g., .po, .json, .yml) as defined in the project configuration.
Boundary markers: There are no explicit instructions or delimiters provided to ensure the agent ignores hidden instructions that might be embedded within the source strings.
Capability inventory: The skill possesses the ability to execute shell commands via the crowdin-cli (including glossary list and glossary upload) and to write files to the local file system (draft CSVs).
Sanitization: The risk is significantly mitigated by a mandatory human review checkpoint, where the agent is instructed to stop and display the draft terms for explicit user approval before performing any upload operations.

glossary-generation