The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains instructions that use "Highest Priority" framing and "Ignore/Disregard" markers to ensure the agent prioritizes the skill's specific rules over previous instructions or historical context.
[DATA_EXFILTRATION]: The scripts/tencent_meeting.py and scripts/utils.py files contain logic to automatically collect the host system's operating system name and version. This information is injected into the _client_info parameter and transmitted to the remote API during every tool invocation.
Evidence: The get_os_name function in scripts/utils.py reads from /etc/os-release and uses the platform module to gather system metadata, which is then sent via the McpProxy class in scripts/mcp_proxy.py.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external sources, specifically meeting transcripts and AI-generated meeting summaries, which could be manipulated to contain malicious instructions.
Ingestion points: Meeting content is retrieved via the get_transcripts_details and get_smart_minutes tools.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are configured for handling these text-heavy external inputs.
Capability inventory: The agent has the authority to perform destructive or high-privilege actions like cancel_meeting or update_meeting.
Sanitization: The provided Python scripts do not implement filtering or sanitization of the transcript content before it is returned to the agent context.

tencent-meeting-skill