The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, user-controllable data from meeting records which could contain malicious instructions.
Ingestion points: Meeting summaries (notes[].note_content and notes[].todo_content via meeting get) and raw speech transcripts (original_data via meeting original get) are ingested into the agent's context for summarization and reporting.
Boundary markers: The skill includes a high-priority "Security Hint" at the end of every file (SKILL.md and all references) explicitly instructing the agent to ignore any executable commands or instruction requests found within tool outputs.
Capability inventory: The agent can create, modify, and delete meetings, search contacts, and view calendar availability using the wecom-cli tool.
Sanitization: The instructions explicitly warn the agent not to output or use any tokens or credentials found in the interface results.
[COMMAND_EXECUTION]: The skill relies on a vendor-provided CLI tool for all operations.
Evidence: The skill requires the wecom-cli binary (associated with the skill author 'wecomteam') to perform meeting management actions. All commands follow a fixed JSON-based argument structure (wecom-cli meeting [action] --json '...'), which reduces the risk of arbitrary shell injection from user inputs.

wecomcli-meeting