The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input describing corporate actions and processes external legal precedents to generate resolutions, creating a surface for indirect instructions to influence the output draft.
Ingestion points: Processes user-provided descriptions of board actions and retrieves precedent resolutions from repository files.
Boundary markers: Implements explicit logic to halt execution and warn the user for 'Major Actions' or when precedents are missing, requiring manual confirmation to proceed.
Capability inventory: The skill is designed to load context from and write draft documents to specific local directories within the user's home directory.
Sanitization: Relies on mandatory professional review instructions and explicit user choice between pre-defined paths rather than automated input filtering.
[DATA_EXFILTRATION]: The skill accesses local files that contain sensitive corporate governance information, including board member lists and resolution histories.
Evidence: Instructions load data from paths such as ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md to retrieve director identities and internal templates.

written-consent