academic-pipeline

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly instructs agents to retrieve and read external reference texts and web sources (e.g., integrity_verification_agent's Phase A/C with WebSearch/DOI checks in SKILL.md and claim_ref_alignment_audit_agent.md's "retrieve full text (api / manual_pdf / paywall / not_found / audit_tool_failure)"), and those third‑party contents are used to make blocking decisions that alter pipeline actions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly dispatches the claim_ref_alignment_audit_agent which at the Stage 4→5 transition retrieves cited full-text from external sources (e.g., DOI/publisher URLs like https://doi.org/10.1186/s41239-026-00585-x) and uses that retrieved text to drive its claim-alignment judgments—i.e., external URLs are fetched at runtime and their content is injected into agent prompts that control decisions.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)