The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for the agent to ingest and process user-provided text through structured form widgets. While this creates a potential surface for indirect prompt injection, the skill includes several defensive patterns to mitigate risks.
Ingestion points: User input enters the agent context through the host-specific elicitation tools described in SKILL.md (ask_followup_questions for Codex and visualize.show_widget for Claude Code).
Boundary markers: The instructions rely on structured platform widgets rather than free-form text blocks, which naturally separates specific inputs, although explicit delimiters for text field content are not defined.
Capability inventory: The skill uses asset-import to handle media assets and triggers tool-based form submissions. It specifically restricts the agent from returning raw bytes or local system paths.
Sanitization: The skill explicitly instructs the agent to verify every required answer after submission and before continuing the workflow, and mandates the use of host-supported safe media renderers instead of raw HTML or bare URLs.

widget-forms