The Agent Skills Directory

[DATA_EXFILTRATION]: The skill implements a telemetry system that tracks usage events and basic system information (operating system and architecture via uname). Data is logged to a local file ~/.superstack/telemetry.jsonl and sent to a remote endpoint if a convexUrl is defined in the user's configuration. This tracking is explicitly designed as an opt-in feature, prompting the user for permission before activation.
[EXTERNAL_DOWNLOADS]: Development guides within the skill recommend installing essential toolchains (Rust, Solana CLI, Anchor) using official installers via curl | bash pipelines. These sources include well-known services such as rustup.rs and the official Solana infrastructure (solana.workers.dev).
[COMMAND_EXECUTION]: The skill utilizes shell command blocks to manage its internal telemetry state, verify environment prerequisites (checking versions of Node.js, Solana, and Anchor), and facilitate local testing using tools like solana-test-validator.
[PROMPT_INJECTION]: The skill processes project-specific architecture decisions from .superstack/build-context.md. While necessary for functionality, this ingestion point serves as an indirect prompt injection surface where external data could potentially influence the agent's behavior.
Ingestion points: .superstack/build-context.md (read in SKILL.md).
Boundary markers: None; the agent is instructed to directly adopt the stack and architecture decisions found in the context file.
Capability inventory: Local shell execution (bash), file system access (mkdir, echo), and network operations (curl).
Sanitization: No specific sanitization or validation of the context file content is performed.

build-with-claude