The Agent Skills Directory

[PROMPT_INJECTION]: The Step 2 prompt template directly interpolates the {topic} and {time_range} variables into instructions for a background agent. A malicious user could craft a research topic containing instructions to override the agent's goals or access internal context.- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from both the user and external web search results which are then used to influence subsequent outputs and file generation.
Ingestion points: User input for {topic} in SKILL.md; Web search results in Step 2.
Boundary markers: None identified; variables are interpolated directly into the markdown-formatted prompt.
Capability inventory: The skill has access to Read, Write, and WebSearch tools.
Sanitization: No evidence of escaping or validation of user-provided strings before they are processed by the model or written to disk.- [COMMAND_EXECUTION]: The workflow creates directories using a {topic_slug} derived from the user input. If the agent does not strictly sanitize this slug, a malicious topic name could be used for directory traversal, causing files to be written outside of the intended project directory.

research