skills/api7/agent-skills/a6/Gen Agent Trust Hub

a6

Pass

Audited by Gen Agent Trust Hub on Sep 16, 2026

Risk Level: SAFECREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The documentation contains multiple hardcoded example credentials used in setup guides. These include a default APISIX Admin API key (edd1c9f034335f136f87ad84b625c8f1) in references/personas/developer.md, well-known AWS mock keys (AKIAIOSFODNN7EXAMPLE) in references/plugins/ai-content-moderation.md, and various placeholder strings for JWT and HMAC secrets.
  • [INDIRECT_PROMPT_INJECTION]: The skill manages a significant indirect prompt injection surface by configuring plugins that process untrusted data.
  • Ingestion points: The ai-proxy and ai-content-moderation plugins (references/plugins/ai-content-moderation.md, references/plugins/ai-proxy.md) ingest client-supplied prompts and LLM-generated responses.
  • Boundary markers: While the skill suggests using ai-prompt-decorator to inject safety guidelines, the examples do not demonstrate strict data delimitation to prevent instruction override within the processed content.
  • Capability inventory: The agent can use the a6 CLI to perform resource management, modify headers, rewrite URIs, and execute Lua code through the gateway.
  • Sanitization: The skill relies on external moderation services (AWS/Aliyun) which may be bypassed by sophisticated adversarial inputs.
  • [DYNAMIC_EXECUTION]: The skill documents the use of high-privilege dynamic execution features. Specifically, references/plugins/serverless.md describes the execution of inline Lua functions compiled at load time, and references/plugins/ext-plugin.md describes the Plugin Runner architecture which allows APISIX to spawn and manage external subprocesses for Go, Java, and Python logic.
  • [EXTERNAL_DOWNLOADS]: The instructions include downloading and installing external components from well-known sources. references/personas/developer.md provides a 'go install' command for the CLI tool from the vendor's repository (github.com/api7/a6), and references/plugins/prometheus.md fetches dashboard configuration from the official Apache repository (raw.githubusercontent.com/apache/apisix).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 16, 2026, 10:03 AM