The Agent Skills Directory

[COMMAND_EXECUTION]: The skill design relies on the Bash(*) tool to execute fetching logic (Python scripts or shell commands) specified in separate adapter definition files (adapters/trend-sources/*.md). This allows for dynamic execution based on local configuration files. There is a potential path traversal risk if the sources argument is not properly validated when constructing file paths for these adapters.
[EXTERNAL_DOWNLOADS]: The skill utilizes WebFetch to gather trending topics from several public websites (e.g., Reddit, Weibo, Zhihu) and to retrieve content from URLs manually provided by the user. While standard for this type of tool, it involves regular interaction with untrusted external content.
[PROMPT_INJECTION]: As the skill ingests and processes text from various web sources and user pastes for its 'Scoring' phase, it is susceptible to indirect prompt injection. Maliciously crafted content on trend platforms could attempt to influence the agent's behavior during its analysis or when writing to the candidate list.
Ingestion points: Untrusted text retrieved via WebFetch from multiple trend sources and strings pasted by the user into the manual-paste adapter (as seen in SKILL.md, Phase 1-2).
Boundary markers: No specific delimiters or instructions to ignore embedded commands are described in the workflow for handling fetched content.
Capability inventory: The skill possesses Bash(*), Write, Edit, Glob, WebFetch, and Skill capabilities, which could be abused if an injection is successful (as seen in the allowed-tools section).
Sanitization: The instructions do not mention any sanitization, filtering, or validation steps for the fetched data before it is passed to the scoring logic (as seen in Phase 4).

cheat-trends