The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file SKILL.md contains a hardcoded sensitive credential labeled as a "Real API key" (7a4b2ef1a7b209016ce2670b6b01fc3066a3553bfae0b0745aaf44266dba724c) for the am_server backend.
[REMOTE_CODE_EXECUTION]: The skill provides instructions and API endpoints (e.g., wss://api.sprites.dev/v1/sprites/{name}/exec) specifically designed to run arbitrary code and shell commands on remote microVMs.
[COMMAND_EXECUTION]: The reference document references/api-workflows.md details high-privilege command patterns including systemctl restart, pkill, and git pull to be executed on remote systems.
[PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface by ingesting data from external files and logs into the agent's context.
Ingestion points: Filesystem API (GET /v1/sprites/{name}/fs/{path}) and log tailing via Exec API as described in references/api-workflows.md.
Boundary markers: None identified; instructions do not specify delimiters or warnings to ignore embedded content.
Capability inventory: Arbitrary shell execution (/exec), filesystem management (/fs), and background service control.
Sanitization: No evidence of sanitization, escaping, or validation of the retrieved remote content before it enters the prompt context.

sprites