competitor-analysis

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

• Third-party content exposure detected (medium risk: 0.30). In the required runtime workflow, the main agent uses Browserbase discovery (browse cloud search) and then fetches and reads outsider-authored web content (candidate homepage/title/hero text) via browse cloud fetch in the Gate step (scripts/gate_candidates.mjs), which directly ingests untrusted free text from user-controlled/outsider-authored sources.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The skill explicitly instructs the user/agent to permanently approve and bypass interactive tool permission prompts by editing ~/.claude/settings.json and to batch file-write and command approvals, which is a directive to circumvent security prompts and thus risks compromising the host state.