The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill handles natural language prompts and processes model-generated artifacts from the Intent Agent. * Ingestion points: The skill reads JSON-RPC responses containing artifact text from the Intent Agent in task.json. * Boundary markers: Documentation in SKILL.md and intent-reference.md contains explicit instructions to treat evaluations, reasons, and obligations as data for processing rather than instructions for the agent. * Capability inventory: The skill utilizes curl and jq within the scripts/intent.sh script to interact with the Intent Agent service. * Sanitization: Input prompts are escaped using jq when building requests, and output is parsed as structured JSON.
[COMMAND_EXECUTION]: The skill includes a bash script (scripts/intent.sh) that executes curl and jq commands to communicate with the Intent Agent API. The script uses bash arrays and jq-based JSON construction to handle variables safely.
[DATA_EXFILTRATION]: The skill performs network operations to external endpoints, including IndyKite vendor domains (indykite.com, indykite.io) and well-known services like Google Gemini for intent extraction. It handles authentication credentials such as bearer tokens and API keys via environment variables.

indykite-intent-agent