subagent-driven-development

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill explicitly instructs that "exact values" (numbers, magic strings, signatures) be placed in brief files and tells implementer subagents to "use verbatim" and that everything pasted into dispatch prompts and subagent outputs remains in the session context, which forces the LLM to receive and potentially emit secret or sensitive strings verbatim.

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

• Third-party content exposure detected (medium risk: 0.30). The workflow’s runtime LLM inputs include the task brief content extracted from PLAN_FILE via scripts/task-brief (used to dispatch implementer subagents), meaning outsider-authored free text placed into the plan is ingested directly by the agent without selecting a specific item beyond the task number.