The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents how to build automations that ingest untrusted data and process it via AI agents with access to powerful tools. This creates an indirect prompt injection surface.
Ingestion points: SELF_SERVE_TRIGGER (user-provided form data) and EVENT_TRIGGER (external entity changes from the context lake) in references/node-types.md and example files like assets/event-driven-ai-incident-response.json.
Boundary markers: The documentation and examples (e.g., assets/event-driven-ai-incident-response.json) show direct interpolation of trigger data into AI prompts (e.g., {{ .outputs.trigger.diff.after.title }}) without explicit delimiters or instructions to ignore embedded commands.
Capability inventory: The workflow nodes include WEBHOOK (arbitrary HTTP requests), INTEGRATION_ACTION (GitHub/GitLab/Azure DevOps operations), CURSOR_AGENT (code modification), and UPSERT_ENTITY (modifying Port's internal data), as detailed in references/node-types.md.
Sanitization: No mention of input sanitization or JQ-based filtering for safety is provided in the references/data-flow.md or references/node-types.md instructions.

port-workflows