The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file contains bash script blocks in the Preamble and Telemetry sections designed to be executed automatically at the start and end of the skill's lifecycle. These scripts interact with the local filesystem and perform network operations.
[REMOTE_CODE_EXECUTION]: Documentation within the references folder (e.g., references/why-solana-for-builders.md) explicitly provides commands for users to pipe remote shell scripts directly into bash (e.g., curl -fsSL ... | bash). This practice is inherently risky as it allows for arbitrary code execution from external domains.
[DATA_EXFILTRATION]: The automated preamble script extracts system hardware information using uname and sends usage metrics to a remote endpoint (defined as convexUrl in the user's configuration) via curl POST requests.
[INDIRECT_PROMPT_INJECTION]: The skill captures user input during an interview process and writes it as a JSON object to a shared context file (.superstack/idea-context.md). Because this file is intended to calibrate the behavior of subsequent skills, it represents an attack surface for multi-step prompt injection if the user provides malicious input.
[PERSISTENCE]: The skill manages its own state and tracking by creating and modifying hidden files and directories in the user's home path (~/.superstack), ensuring that telemetry settings and session data persist across multiple interactions.

solana-beginner