The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local pre-compiled binaries (./spark/tools/memory and ./spark/tools/tavily-search) to perform database operations and network searches. Bundled binaries are opaque, meaning their internal behavior cannot be audited or verified, which poses a risk as they operate with the same privileges as the agent.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to the Tavily search service to gather market research data. This involves sending user-related keywords to an external API and requires the user to provide a TAVILY_API_KEY in the environment.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes content from external search results without proper sanitization or boundary markers.
Ingestion points: Step 4 parses JSON output from the tavily-search tool which contains content from arbitrary third-party websites.
Boundary markers: Absent. The instructions do not include delimiters or specific guidance to the agent to ignore any instructions found within the retrieved search data.
Capability inventory: The skill can execute shell commands via the bundled binaries and maintain state in a local SQLite database (spark/memory.db).
Sanitization: Absent. There is no evidence of validation or filtering for the external content before it is processed by the agent to generate ideas.

spark