acontext-chatbot-integration

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The integration explicitly fetches and mounts user-provided skills (references/skill-system.md and SKILL.md) via APIs like client.skills.list_catalog(), client.skills.get_file() and SKILL_TOOLS/SANDBOX_TOOLS (skills are read into the agent's system/context and mounted into sandboxes at /skills/), meaning untrusted, third‑party skill files can be ingested and influence the agent's tool use and behavior.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The integration calls the Acontext API at runtime (baseUrl https://api.acontext.com/api/v1) to fetch mounted skill files (SKILL.md) which are injected into system prompts via SKILL_TOOLS and to create/exec sandboxes (execCommand), so remote content and commands fetched from that URL can directly control prompts or execute code.