The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute a local script scripts/benchmark_fetch.py using python3 to perform search, cost estimation, list retrieval, and media downloading tasks.
[EXTERNAL_DOWNLOADS]: The skill initiates network requests to tikhub.io to fetch social media data from platforms including Douyin, Xiaohongshu, and WeChat Channels. It also suggests installing external software such as whisper-cpp via brew for transcription.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content, specifically 'transcripts' and 'cover text' from social media accounts, which are ingested into the agent's context for analysis.
Ingestion points: Content is saved into transcript.md and meta.md from external social media posts.
Boundary markers: No explicit boundary markers or 'ignore' instructions are defined for processing these files.
Capability inventory: The agent reads these files to generate a 'teardown and roadmap' (99-拆解与方案.md).
Sanitization: There is no evidence of sanitization or filtering applied to the external text before it is analyzed by the LLM.
[PERSISTENCE]: The instructions direct the user to persist sensitive credentials by appending the TIKHUB_API_KEY to the ~/.env file using shell commands.

zmm-benchmark