news-briefing-research
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes news articles and metadata from external sources such as GDELT, BBC, and CNN. This untrusted content could potentially contain instructions designed to influence the agent's behavior.
- Ingestion points: Content is ingested via the
scripts/crawlora.shhelper script in theSKILL.mdresearch workflow. - Boundary markers:
SKILL.mdincludes specific instructions to "label unverified allegations, corrections, missing context" and "clearly separated reporting from inference." - Capability inventory: The skill performs network GET requests via
curland processes JSON output usingjqas documented inSKILL.md. - Sanitization:
scripts/crawlora.shimplements a strict path whitelist for API routes, validates the API key format to prevent shell injection, and uses--data-urlencodeto safely handle all query parameters. - [EXTERNAL_DOWNLOADS]: The skill fetches news data from the vendor's official Crawlora API at
api.crawlora.net. - [COMMAND_EXECUTION]: Uses a local bash script (
scripts/crawlora.sh) to wrapcurlcalls. The script includes robust safety checks, including the use of temporary files with restricted permissions (chmod 600) for handling API keys andumask 077to ensure secure file creation.
Audit Metadata