The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates processing output from lark-cli, which can contain content from external Lark/Feishu documents, wiki nodes, and API responses, creating a potential surface for indirect prompt injection.
Ingestion points: Structured JSON output from lark-cli (stdout/stderr) containing remote data (e.g., lark-cli wiki +node-get).
Boundary markers: The instructions lack specific boundary delimiters for the agent when interpreting untrusted data fields within the JSON output envelopes.
Capability inventory: The agent can perform authentication tasks, manage cloud storage resources (drive), and read/write workspace content.
Sanitization: The skill provides excellent sanitization guidance, including enforcing relative path validation and advising the use of argument arrays (argv) instead of shell concatenation to prevent command injection.
[COMMAND_EXECUTION]: The skill defines clear patterns for executing the lark-cli binary to manage Lark resources.
Evidence: Explicit instructions for lark-cli auth, lark-cli config, lark-cli drive, and lark-cli wiki commands.
Security Protocols: Implements a structured 'high-risk-write' confirmation protocol (exit code 10) requiring explicit user consent before proceeding with dangerous operations.

lark-shared