The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of tweets scraped from x.com. This creates a surface for indirect prompt injection attacks where malicious content within a tweet could attempt to influence the agent's behavior during the repurposing process.\n
Ingestion points: Tweet content is ingested through the XActions.scan() function in src/contentRepurposer.js and the x_get_tweets tool as defined in the MCP Workflow section of SKILL.md.\n
Boundary markers: The instructions do not define clear delimiters (e.g., XML tags or unique markers) or include 'ignore embedded instructions' warnings to prevent the agent from following commands embedded in the tweet text.\n
Capability inventory: The skill identifies high-engagement content and generates variations. It also includes the src/autoPlugReplies.js script, which implies a capability to post automated replies to the platform, potentially creating a vector for unauthorized posting if influenced by injection.\n
Sanitization: No evidence of input sanitization, content filtering, or validation of the ingested tweet data is provided in the skill instructions or workflow descriptions.

content-repurposing