The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns and examples for building AI agents (e.g., in references/ai/agent-tool-loop.md) that ingest untrusted data which can influence the agent's logic.
Ingestion points: Untrusted data enters the agent context through event payloads, such as event.data.message in the provided ReAct loop examples.
Boundary markers: The documented code samples lack explicit delimiters or instructions to the LLM to ignore potentially malicious instructions embedded in the input data.
Capability inventory: The described agent workflows possess significant capabilities, including tool execution and data processing, which can be manipulated via prompt injection.
Sanitization: There is no evidence of sanitization or validation for external natural language content before it is interpolated into LLM prompts in the examples provided.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions for installing platform tools via remote shell scripts.
Evidence: The skill references curl -sSfL https://cli.inngest.com/install.sh | sh for CLI installation in references/platform/dev-server.md and scripts/setup.md.
Note: These downloads originate from the official Inngest domain (inngest.com), representing standard and expected behavior for this platform's reference skill.
[COMMAND_EXECUTION]: The skill extensively documents and encourages the use of various command-line interface tools for development, testing, and deployment.
Evidence: Numerous files in the scripts/ and references/platform/ directories describe the use of inngest dev, npx inngest-cli, and package managers (npm, pip) to interact with the system.

inngest