The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes bash scripts in the preamble and postamble sections to initialize telemetry settings, update local configuration files, and log execution details. These scripts utilize shell utilities like grep, sed, and cut to parse and manipulate configuration data within the user home directory.
[DATA_EXFILTRATION]: The skill transmits telemetry data, including the skill identifier, session ID, and system environment details (OS and hardware architecture), to a remote URL defined in the user's local configuration file. Evidence: SKILL.md uses curl -s -X POST "$_CONVEX_URL/api/mutation" to send JSON payloads to an external service.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze data from the DefiLlama API, creating a surface for indirect prompt injection if the external data source provides malicious content. 1. Ingestion points: The workflow fetches data from api.llama.fi, yields.llama.fi, and stablecoins.llama.fi via direct API calls. 2. Boundary markers: The instructions do not define specific delimiters or instructions to ignore potential injections within the fetched API data. 3. Capability inventory: The skill performs shell script execution, makes network requests, and writes structured JSON data to .superstack/idea-context.md. 4. Sanitization: No explicit validation or filtering logic is provided for the content retrieved from the external API before it is processed by the agent.

defillama-research