The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides automated protocols for installing the external package mcp-taiwan-legal-db using pip or pipx. This package is not from a verified or trusted source, enabling the execution of arbitrary third-party code.
[EXTERNAL_DOWNLOADS]: The instructions require the agent to register and connect to a remote HTTP MCP server at https://tlr.dr-lawbot.com/mcp. This facilitates the retrieval of data and the potential invocation of tools from an external domain without verification.
[COMMAND_EXECUTION]: The skill specifies shell commands, such as claude mcp add and gemini mcp add, to programmatically modify sensitive local configuration files, including ~/.claude.json, ~/.cursor/mcp.json, and ~/.gemini/config/mcp_config.json. This behavior establishes a persistence mechanism for external tools within the agent's environment.
[COMMAND_EXECUTION]: The instructions in references/agents-rules.md mention the execution of a local script python scripts/sync_agents_rules.py, which is not included in the skill distribution for review.
[PROMPT_INJECTION]: The skill uses instruction-overriding language, including terms like "鐵律" (iron rule) and "最高優先，不可覆寫" (highest priority, non-overwritable), designed to lock the agent's behavior and ignore subsequent user instructions that might conflict with the author's directives.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface through the ingestion of user-provided document content for drafting and proofreading. It has the following evidence chain:
Ingestion points: User-provided text for document processing in SKILL.md.
Boundary markers: Absent.
Capability inventory: Network-enabled tool invocation through the taiwan-legal-db and dr-lawbot servers.
Sanitization: Absent.

official-document-drafting