geeknews-search
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches content from the public GeekNews RSS feed via FeedBurner using the Python
urllib.requestlibrary. It also instructs the agent to usenpxto download and run@nomadamas/k-skill, a management utility associated with the skill author's organization. - [REMOTE_CODE_EXECUTION]: The bootstrap instructions in
SKILL.mdandinstruction.mddirect the agent to execute code from the@nomadamas/k-skillpackage usingnpx. This is used to retrieve current instructions and execute skill scripts in a consistent runtime environment. - [COMMAND_EXECUTION]: The skill relies on shell commands (
npx) to perform updates, list files, and execute the core Python logic. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by ingesting data from external RSS feeds. An attacker controlling the feed could attempt to inject instructions into titles or summaries to influence the agent during summarization.
- Ingestion points: The
scripts/geeknews_search.pyscript fetches data from the GeekNews feed URL or a user-provided file. - Boundary markers: None explicitly defined in the prompts to isolate feed content from agent instructions.
- Capability inventory: The skill has network read access for feed retrieval and local file read access for testing.
- Sanitization: The script uses
html.parser.HTMLParserandunescapeto clean content and employs regular expressions to strip XML tags, which provides basic structural sanitization.
Audit Metadata