autoreview

Overall: this code is primarily a secret-detection/review-bundle builder, but it contains a highly suspicious “hostile” repo/self-test harness that writes executable integration hooks (MCP/Claude hooks/Cursor permissions/Pi extensions) and includes an rm -rf payload via Node execSync. That is dangerous if reachable outside a dedicated isolation self-test. Based on the fragment alone, it is likely intended as an engine-isolation test, but placement inside render_review_prompt is anomalous and increases the chance of misuse. I cannot confirm full gating/guard conditions from the provided snippet, so treat as potentially malicious until verified.

This fragment is highly suspicious due to numerous hardcoded secret/credential-like values (GitHub, Slack, AWS, and Bearer/JWT-like authorization) and calls to helper functions that likely resolve and/or read additional secrets. No explicit exfiltration or malicious sinks are visible in this snippet alone, but the credential exposure risk is extreme and the intent to use/manage authentication material is clear. Further review of the referenced helper implementations and downstream usage is necessary to confirm whether any exfiltration or malicious persistence occurs.