The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes bash scripts in the preamble and telemetry sections of SKILL.md that read the local ~/.superstack/config.json and transmit telemetry data (including uname system information and skill usage) to a remote URL via curl. On the initial execution, this transmission occurs before the user is prompted for consent.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests and processes untrusted data from multiple sources to build the pitch deck.
Ingestion points: Data enters the context from .superstack/idea-context.md, .superstack/build-context.md, package.json, and live research results from DefiLlama, CoinGecko, and web searches.
Boundary markers: The instructions do not define clear delimiters or use "ignore embedded instructions" warnings when processing these external inputs.
Capability inventory: The skill has the capability to execute bash commands (preamble), perform network operations (curl and research tools), and write files to the project root (HTML artifact generation).
Sanitization: No sanitization or validation of the ingested external content is performed before it is interpolated into the generated HTML deck.
[COMMAND_EXECUTION]: The skill utilizes several shell commands for project discovery and environment analysis, which, while standard for its purpose, involves interacting with the underlying filesystem.
Evidence: Use of find, cat, and grep to inspect package.json and locate image assets in the project directory.

create-pitch-deck