proxelar

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

• Third-party content exposure detected (high risk: 1.00). Yes — the skill routes arbitrary HTTP/HTTPS traffic through a local MITM proxy and explicitly exposes that untrusted third‑party traffic to agents via the Web GUI and log/capture scripts (see "Agent-Browser Bridge" / SKILL.md and README plus scripts/capture_for_eval.lua and log_to_jsonl.lua), meaning the agent is expected to read/interprete external web/API content that can materially influence subsequent actions.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs trusting a generated CA via a sudo command that modifies the system keychain (and advises binding to 0.0.0.0), which asks the agent to perform privileged, persistent changes to the host.