The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches and runs the spets package from the npm registry using the npx command during skill initialization.
[REMOTE_CODE_EXECUTION]: The skill architecture defines a loop where the agent executes instructions and commands provided by an external orchestrator, effectively allowing remote control of the agent's behavior.
[COMMAND_EXECUTION]: Executes shell commands via npx and is instructed to run commands found in the onComplete field of the orchestrator's JSON response.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes and follows instructions from an external tool without validation.
Ingestion points: JSON output returned from the npx spets command in SKILL.md.
Boundary markers: No delimiters or ignore-instructions warnings are used for the external content.
Capability inventory: Shell execution (npx), file system interaction, and general agent capabilities.
Sanitization: No validation or sanitization of the remote prompt or onComplete fields before execution.

spets