lab-review

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill requires reading local JSONL session files and explicitly instructs the agent to quote "real string[s] from JSONL" into the review, which can force emitting any API keys/passwords or other secrets present in those transcripts verbatim.

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

• Third-party content exposure detected (medium risk: 0.65). SKILL.md specifies reading local Claude Code session transcripts from ~/.claude/projects/<cwd-slug>/<session-id>.jsonl (JSONL event message.content), which at runtime are outsider-authored user/assistant messages from other participants (or at least non-operating-user contributions) rather than only the operating user’s own text.