security-audit

name	security-audit
description	Use when reviewing Nevermind for security risk, threat modeling, secure-code review, dependency/config exposure, secrets, OWASP-style web/API issues, Electron attack surface, auth/session/token/billing abuse, or any request for a penetration-test/security report. Explore the repo before conclusions; this is for authorized review of this app only.

Security Audit

Use this skill for defensive security work on Nevermind. Produce evidence-backed findings, not generic advice.

First steps

1. Establish scope from the repo. Identify the desktop app, backend, deployed surfaces, auth providers, data stores, IPC boundaries, extension APIs, and privileged OS actions before judging risk.
2. Check local guidance. Read AGENTS.md, then the relevant project skill files. For Electron-specific work, read ../electron-best-practices/SKILL.md and linked files.
3. Map trust boundaries. Renderer ↔ preload ↔ main, extensions ↔ host API, user content ↔ extension views, desktop ↔ backend, backend ↔ upstream AI providers, auth sessions ↔ API tokens, admin ↔ user.
4. Audit high-impact paths first. Auth, tokens, admin APIs, proxy/billing, custom protocols, shell/file/URL actions, IPC handlers, extension permissions, secrets/config, update/release pipeline.
5. Verify findings. Prefer code references, concrete reproduction steps, safe local tests, and exact commands over speculation.

Nevermind security surfaces

• Electron main/preload/renderer: src/electron/main.ts, src/electron/preload.ts, src/electron/palette-window.ts, src/preload-api.ts.
• Extension host and API: src/resources/nevermind-extension-api.d.ts, src/electron/main.ts, src/extension-view.tsx, internal extensions under src/resources/.
• Desktop auth/model proxy integration: src/electron/nevermind-auth.ts, src/electron/ai.ts.
• Astro backend routes: backend/src/pages/api/**, especially auth, device auth, admin, tokens, and /api/v1/** proxy routes.
• Backend libraries: backend/src/lib/workos.ts, tokens.ts, admin.ts, proxy.ts, ratelimit.ts, cron-auth.ts, audit.ts, settings.ts.
• Persistence/config: backend/src/db/schema.ts, Drizzle migrations, electron-builder.yml, backend/vercel.json, env examples.

Audit checklist

• Auth/session: redirect handling, sealed cookie settings, CSRF on cookie-authenticated mutations, logout, device-code approval/exchange, role escalation.
• API tokens: generation entropy, prefix leakage, hashing, revocation, last-use writes, bearer extraction, rate limits, token exposure in logs/UI.
• Authorization: admin gates, object ownership checks, extension permission checks, IPC caller assumptions, path/URL action ownership.
• Electron hardening: context isolation, sandbox, node integration, navigation/window-open policy, permission handlers, CSP/custom protocol behavior, preload API input validation.
• Extension isolation: least-privilege permissions, shell/system actions, file and app helpers, iframe sandbox, untrusted HTML/media, clipboard and AI permissions.
• Backend proxy/billing: transparent auth swap, forwarded headers, request size/token limits, upstream URL construction, streaming parsing, abuse/rate limits, billing failure modes.
• Input/output safety: SSR/script injection, DOM insertion, markdown/rendered content, URL/path validation, shell argument handling, log redaction.
• Secrets/config: committed env files, Vercel env exposure, Sentry DSNs, WorkOS/Redis/Neon keys, release signing material, local auth storage.
• Supply chain: dependency vulnerabilities, Electron version posture, packaged resources, updater/release artifact trust, native modules.

Report format

For a full report, include:

1. Scope and methodology.
2. Architecture/trust-boundary summary.
3. Executive summary with severity counts.
4. Findings sorted by severity, each with: title, severity, affected files/routes, evidence, impact, reproduction or exploit sketch, recommended fix, verification command/test.
5. Positive controls already present.
6. Residual risks and follow-up hardening roadmap.

Do not print secrets, live tokens, cookies, raw user data, or production payloads. Redact values while preserving enough shape to prove the issue.