The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill performs research by fetching and synthesizing content from external sources such as Reddit, GitHub, and market reports using WebSearch and WebFetch in SKILL.md.
Ingestion points: Data enters the agent's context through external web searches and fetching URLs from community platforms.
Boundary markers: There are no explicit markers or instructions to treat external data as untrusted or to ignore embedded instructions within the fetched content.
Capability inventory: The skill is configured with Bash and Write tools, which could potentially be misused if the agent inadvertently executes instructions found during research.
Sanitization: The instructions do not include steps to sanitize, filter, or escape data retrieved from external sources before it is processed by the agent team.
[COMMAND_EXECUTION]: The allowed-tools field in the frontmatter grants the skill access to Bash. While intended for research-related tasks, this capability increases the potential impact of indirect prompt injection attacks originating from untrusted web content.
[EXTERNAL_DOWNLOADS]: The 'Common Issues' section recommends that users set up a self-hosted search adapter using a repository hosted on the author's GitHub account (fortunto2/searxng-docker-tavily-adapter) to improve search results.

solo-swarm