toil-mining

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). This skill explicitly extracts and emits raw "request text" and representative evidence from supplied session JSONL without modifying sources (and tells you to include checked/not-checked sources), which forces the agent to reproduce any embedded secrets (API keys, tokens, passwords) verbatim in its output.

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

• Third-party content exposure detected (medium risk: 0.30). The required runtime path reads caller-supplied Codex JSONL session files (explicit paths) and ingests the payload.message free text from event_msg/user_message records via scripts/recent_human.py (extract() → normalize_request()), so an outsider who can author those JSONL inputs can supply indirect prompt-injection text.