The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill fetches content from untrusted external URLs and processes it to generate reports, creating a vector for embedded malicious instructions.
Ingestion points: External content is retrieved using the WebFetch tool as described in Step 2 of SKILL.md.
Boundary markers: The instructions do not specify the use of delimiters or isolation markers (e.g., XML tags) to separate the untrusted article content from the agent's internal analysis prompts.
Capability inventory: The skill performs file creation and write operations to the reports/articles/ and logs/ directories based on processed content.
Sanitization: There is no evidence of sanitization or filtering of the fetched external content before it is interpolated into the analysis prompt.
[EXTERNAL_DOWNLOADS]: The skill metadata contains an installation command that downloads the skill definition from a remote repository.
Evidence: The install_command in metadata.json uses curl to fetch the SKILL.md file from raw.githubusercontent.com.

article-analysis