The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill relies on npx to download and execute the @nomadamas/k-skill package from the NPM registry to retrieve instructions and execute its core scripts. This package is a vendor-owned resource used for skill management.
[COMMAND_EXECUTION]: The Python scripts (_naver_http.py) include an optional --insecure flag that disables SSL certificate verification (ssl.CERT_NONE) for network requests. Although scoped to Naver domains, this feature could allow for person-in-the-middle attacks if enabled in insecure network environments.
[EXTERNAL_DOWNLOADS]: The skill performs automated network requests to well-known Naver domains (.naver.com, .pstatic.net) to retrieve blog posts and download images. These operations are essential to the skill's functionality.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external Naver blog posts, which could contain malicious instructions designed to influence the AI agent's behavior.
Ingestion points: Search result snippets and full blog post content extracted in scripts/naver_search.py and scripts/naver_read.py.
Boundary markers: Absent. The external content is presented to the agent as raw text within JSON structures without explicit isolation or delimiters.
Capability inventory: The skill can write files to the local system and perform network requests to Naver domains via the scripts/naver_download_images.py script.
Sanitization: The scripts perform HTML tag stripping but do not analyze or filter the semantic content of the extracted text for potential prompt injection attacks.

naver-blog-research