ai-coding-agents-provider-runtime
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADS
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides numerous references to official documentation and implementation source code from trusted organizations, including OpenAI, Anthropic, Google, and the Agentic AI Foundation (Linux Foundation).
- Evidence: Links to platform.openai.com, docs.anthropic.com, ai.google.dev, and GitHub repositories for anthropics/claude-code and openai/codex are included in data/sources.json and various reference files.
- [INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is the design of systems that ingest and process untrusted data from AI model providers, which inherently involves an indirect prompt injection attack surface.
- Ingestion points: The runtime event model in SKILL.md processes tokens, tool calls, and errors from external providers.
- Boundary markers: The skill instructs developers to implement a stable internal event model to normalize provider outputs before they reach the rest of the runtime.
- Capability inventory: The guidelines cover tool calling, vision support, and structured output handling across multiple providers.
- Sanitization: Instructions emphasize the normalization and validation of tool-call arguments and structured outputs to prevent malformed payloads from affecting downstream logic.
Audit Metadata