HARPA AI: Web Automation with ChatGPT, Claude, Gemini, Grok
eanggfilgoajaocelnaflolkadkeghjp
Risk Score
5.66
Risk Level:
Medium
Recommendation:
🟡 MEDIUM RISK — review
Top Risks
- browsingData+cookies+scripting+<all_urls>: can read, modify, exfil all site data and session credentials.
- Privacy policy admits data_collection+third_party_sharing but scope_extension==false — no extension-specific disclosure.
- brand_mention impersonation (ChatGPT, Claude, Gemini) without verified publisher status inflates reputation risk.
- nj-engine.js executes new Function(userInput) — dynamic code eval surface in automation engine.
- AI extension processing page content across all URLs creates broad passive data-collection surface.
Evidence
- broad_host_permissions manifest host_permissions=[*://*/*] + content_scripts on *://*/* — all sites reachable.
- high_capability_combo manifest browsingData + cookies + scripting + <all_urls>: can read/modify/exfiltrate full browsing state.
- privacy_policy_unscoped api Policy fetched; data_collection+third_party_sharing=true but scope_extension=false → +10.0 privacy (v3.5-D).
- brand_impersonation store is_impersonation=true (Claude, Gemini, ChatGPT); not verified_publisher, is_featured_by_google=true → +1.0.
- dynamic_code_exec crx new Function(e)() in nj-engine.js executes user/page-supplied strings — code-quality risk.
- ai_page_content store AI extension processing page content across all URLs; +2.5 webstore AI signal.
- no_developer_name store developer_name is empty string; +1.0 reputation penalty.
- featured_by_google store is_featured_by_google=true provides partial reputation mitigation (-2.0 featured badge).
Permissions Breakdown
- alarms low Schedules background tasks; low standalone risk.
- background low Persistent background page; expected for AI automation tool.
- browsingData high Can delete cookies, cache, history — broad destructive capability.
- cookies high Read/write all cookies across all sites via *://*/*.
- declarativeNetRequest medium Can block/redirect network requests; moderate risk.
- notifications low Display notifications; minimal risk.
- tabs medium Access URL, title, tab state across all tabs.
- storage low Local/sync storage; low risk.
- offscreen low Offscreen document API; contained risk.
- scripting high Inject arbitrary scripts into *://*/* pages — critical capability.
- contextMenus low Adds right-click menu items; minimal risk.
- sidePanel low UI side panel; minimal risk.
- *://*/* high Broad host access covering all URLs; amplifies every other permission.
Pillar Scores
Permissions7.50
Reputation5.50
Network3.50
Webstore5.50
Maintenance0.00
Privacy10.00
Code Quality5.00
CVE Exposure0.00
Scoring History
| v3.6 | 5.66 | Medium | review | 2026-06-16 |
| v3.4-rev | 6.04 | High | review | 2026-06-15 |
Bookkeeping
Rubric v3.6
Scored at 2026-06-16 07:29
Listing SHA
17caa1f43894…
Force block
— not fired
Score recovered
no
Elapsed
33.6s