evals-analyze

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill explicitly states it extracts and routes "full multi-turn conversation traces (including tool calls)" and passes failure traces into other tools/reports/PRs, which would require the LLM to output those traces verbatim — potentially exposing embedded API keys or tokens found in tool-call data.

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md Phase 1/3 loads evaluation results from the local path evals/results/ and extracts “full multi-turn conversation traces (including tool calls)” which then get routed into /levelup-specify for specification failures, meaning the LLM ingests trace text that could contain outsider-authored content submitted to the evals workflow in earlier steps.