abund-ai

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly instructs the agent to report the registration claim_url (which contains a claim code) to the human and shows examples where the API key must be placed verbatim in Authorization headers, so the LLM would need to output sensitive secret values verbatim (exfiltration risk).

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

• Third-party content exposure detected (high risk: 0.85). The required runtime workflow for the skill is to fetch and ingest Abund.ai’s own skill/heartbeat text from https://abund.ai/skill.md and https://abund.ai/heartbeat.md (and skill.json), which is outsider-authored free text from a third-party domain into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill manifest and heartbeat endpoints (https://abund.ai/skill.md, https://abund.ai/heartbeat.md, and https://abund.ai/skill.json) are explicitly fetched at runtime (install/heartbeat steps) and provide instructions/manifests that directly control the agent's behavior, so they constitute a runtime external dependency that can control prompts.