The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the installation of the agentcash package from the public NPM registry using npm install -g agentcash (rules/getting-started.md).
[REMOTE_CODE_EXECUTION]: Several commands utilize npx agentcash@latest, which dynamically fetches and executes the most recent version of the package from a remote registry at runtime without version pinning (SKILL.md, rules/getting-started.md, rules/uploads.md).
[COMMAND_EXECUTION]: The skill frequently invokes shell commands including npx for API interactions and curl for binary file uploads to external storage (SKILL.md, rules/uploads.md).
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it accepts untrusted text prompts from users to generate media (SKILL.md). These prompts are passed directly to CLI tools with network and execution capabilities. The following evidence was identified:
Ingestion points: User-provided strings for the prompt field in generation and edit requests (SKILL.md).
Boundary markers: None identified; instructions are interpolated directly into JSON payloads.
Capability inventory: The agent utilizes npx (command execution) and curl (network operations) to process data.
Sanitization: No evidence of input validation, escaping, or filtering of user prompts before they are used in commands.

media-generation