suede-social
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a social listening workflow that retrieves content from external, untrusted sources (Reddit, Hacker News, Bluesky, and RSS feeds) to generate engagement drafts. This creates a surface where instructions embedded in external posts could influence the agent's behavior. 1. Ingestion points: External APIs and RSS feeds via curl commands documented in references/listening.md. 2. Boundary markers: The skill includes a 'Boundaries' section in SKILL.md and a human 'Approval gate' in references/listening.md requiring review before action. 3. Capability inventory: Shell command execution (curl, jq, xmllint) and local file access (.agents/product-marketing.md). 4. Sanitization: No explicit sanitization or escaping of ingested content is described before prompt interpolation.
- [COMMAND_EXECUTION]: The references/listening.md file provides shell command recipes using curl, jq, and xmllint for automated data retrieval. While intended for informational purposes, these involve executing shell operations.
- [EXTERNAL_DOWNLOADS]: The skill fetches data from well-known platforms (Reddit, Hacker News, Bluesky) as part of its core research functionality. These are recognized services used for their established data retrieval purposes.
Audit Metadata