orca-eval

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill instructs scanning and extracting user prompts and repeated Bash commands from session logs and to include short quoted evidence in reports, but gives no redaction rules — so any embedded API keys/passwords in logs would be reproduced verbatim.

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

• Third-party content exposure detected (high risk: 0.85). Yes—assets/orca-eval-scan.sh reads outsider-authored *.jsonl session logs from ~/.claude/projects/<project-slug>/*.jsonl, extracts user message text (free-form prompts) and tool error strings, and prints them into the runtime output that the agent then “Distill/Report” into LLM context.