The Agent Skills Directory

[SAFE]: The skill follows a structured information-gathering and documentation workflow for legal and HR project management.
[DATA_EXFILTRATION]: No network operations (curl, wget, etc.) are present. The skill strictly reads from and writes to a local configuration path (~/.claude/plugins/config/claude-for-legal/employment-legal/) to maintain state across sessions, which is standard behavior for its described purpose.
[PROMPT_INJECTION]: While the skill ingests user input regarding country names and hiring roles (Indirect Prompt Injection surface), it interpolates this data into highly structured YAML and Markdown templates. The risk is minimized as the skill does not interpret this user data as instructions and does not execute it in a shell or code interpreter.
Ingestion points: User responses to Step 1 intake questions (country, roles, headcount).
Boundary markers: Natural boundaries are used via Markdown headers and YAML key-value structures.
Capability inventory: Limited to file system writes of text/YAML data to a specific local directory (Step 5).
Sanitization: No explicit sanitization is noted, but the risk is low due to the lack of executable capabilities involving the ingested data.
[REMOTE_CODE_EXECUTION]: The skill contains no code or instructions to download and execute external scripts. It relies entirely on structured prompt instructions and local file I/O.

international-expansion