agent-experience

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill explicitly asks for/pulls real credential values (paste or auto-discover), writes them into per-agent .env files and permits unabridged tool-inputs in traces (and the orchestrator may run curl with secrets), which creates a high risk that secret values will be handled by the LLM/process and could be output or leaked in traces or tool inputs.

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

• Third-party content exposure detected (high risk: 0.85). Yes: the required workflow has subagents perform runtime WebFetch of outsider-authored public documentation pages (Step 1.25 research + Step 6.25 provenance), and the fetched page text/tool results are then included in the agent’s LLM context via the subagent’s tool-result/prose trace (Step 6.210-214, 6.25.224-237).