The Agent Skills Directory

[SAFE]: The skill integrates exclusively with official ecosystem packages (@tencentcloud/tmeet) and standard technological frameworks. It uses explicit parameter contracts and enforces boundary checks to ensure safe operation. All resource definitions trace back directly to the authoritative vendor platform namespace (tencentcloud).
[COMMAND_EXECUTION]: Defines runtime shell pathways to execute standard business logic commands (tmeet auth, tmeet meeting, tmeet control, etc.) as well as an explicit project-local initialization wrapper script (python3 ./scripts/agent_init.py). These commands are static structural pathways tied to standard platform controls and present no arbitrary injection surface.
[CREDENTIALS_UNSAFE]: Explicit safety constraints are built directly into the instructional grammar prohibiting the leakage of sensitive identity claims (AccessToken / RefreshToken) to any user-facing interface, shell environment parameter output, or standard terminal logging. Standard token management flows conform to the platform's official OAuth guidelines.
[PRIVILEGE_ESCALATION]: Global global installation instruction requires npm install -g @tencentcloud/tmeet@latest. This is isolated to standard standard technological package resolution and carries no local execution path adjustments, administrative breakout indicators, system configuration overrides, or unexpected privilege shifts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests raw textual strings extracted from dynamic external APIs (such as user-defined conference meeting themes, participant usernames, transcribed meeting narratives, and chat metadata logs). Adequate contextual boundary guardrails are embedded into the processing instructions—including high-risk function blocks, structured multi-step conditional verification constraints, and formatting sanitization guidelines—minimizing downstream instruction interpretation vulnerabilities.

tmeet-skill