The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple bash script blocks in the Preamble and Telemetry sections of SKILL.md. These scripts manage local state, create directories (~/.superstack), and log events to local files.
[DATA_EXFILTRATION]: The skill reads system information (platform and architecture) and configuration data, then transmits this data to a remote endpoint using curl. While this is for telemetry and requires user opt-in, it involves sending environment details to an external server.
[PROMPT_INJECTION]: The skill uses a specific instructional pattern to prompt the user for telemetry consent, which influences the agent's initial behavior and overrides the standard workflow until addressed.
[EXTERNAL_DOWNLOADS]: The skill fetches data from external APIs, such as the DefiLlama protocols API, and searches public platforms like GitHub and Twitter to gather competitive intelligence.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to search external, untrusted sources for competitive intelligence, creating a potential attack surface for indirect injection.
Ingestion points: Data is ingested from web research results (Twitter, GitHub, DeFiLlama) and user-provided product ideas.
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands within the ingested search results.
Capability inventory: The agent has the capability to execute shell commands, write files, and perform network requests.
Sanitization: No explicit sanitization or filtering of the external search results is mentioned before the data is processed or written to artifacts.

competitive-landscape