The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from a wide variety of external and untrusted sources, including GitHub Issues, Reddit, Hacker News, and various web pages (L1-L4 sources).
Ingestion points: External data is collected in Step 2 ("资料分层与权威锁定") and Step 3 ("事实抽取与证据卡片") of the methodology described in SKILL.md.
Boundary markers: The instructions lack technical boundary markers or specific prompt-level delimiters to prevent the agent from obeying instructions embedded within the researched materials.
Capability inventory: The skill uses shell commands (tar), file system write operations (~/Downloads/research/), and tool invocations (Search tool, Task tool for agent delegation).
Sanitization: There is no evidence of sanitization or filtering applied to the external content before it is processed into facts and conclusions.
[COMMAND_EXECUTION]: The methodology includes a step to package the research results using a shell command.
Evidence: SKILL.md contains the instruction tar -czvf ~/outcome.tar.gz -C ~/Downloads/research <topic>. The <topic> variable is derived from user input, creating a potential command injection surface if the input is not properly sanitized by the agent before shell execution.
File System Access: The skill creates a nested directory structure and multiple markdown files under the user's ~/Downloads directory.
[DYNAMIC_EXECUTION]: Step 6.5 of the methodology instructs the agent to initiate an independent verification process by starting another agent instance.
Evidence: "启动独立 Agent（Task 工具）" in SKILL.md indicates the dynamic creation and delegation of tasks to a separate agent process via the platform's Task tool.

deep-research