The AI CSO is goal-first. It keeps five standing goals — no open critical exposure (anything on CISA KEV or critical
with a fix is fixed within days), zero secrets in code, verified-only active checks, nothing trusted unvetted
(every new skill, MCP server and package checked first) and every incident with a timeline — and every skill below
exists to move one of them.
Core (topy-ai-cso): connect, first setup (inventory, domain verification, host consent), the CSO's goals, the daily watch.
Vulnerabilities: are we affected by this CVE, dependencies and supply chain, end of life, vetting a package before install.
Code: secrets (rotate first, then remove), unsafe code patterns, a STRIDE threat model and the OWASP ASVS checklist.
Web: domain verification, headers and TLS, a safe scan of a verified site.
Agents: vet a skill or MCP server before it is installed (the App Factory gate), and audit your own MCP server.
Response: incident timelines, a fix plan with fix-risk tiers T1–T4, the weekly security report.
The safety model comes from Maggie Security and does not bend: the CSO never changes a server, firewall or website
(fix and protection plans are for a person to review and apply), never scans a domain you have not verified, reads a
server only after you grant consent yourself in a terminal (an agent cannot), leaves checks of sites you don't control
to a person (never a routine), never uses offensive techniques and never prints a secret. Changes take two calls — a
preview, then your yes.
Install
npx @topy-ai/cso-skills install # this project, for the agents set up here
npx @topy-ai/cso-skills install -g --mcp # your user, plus the MCP server
npx @topy-ai/cso-skills install --category vulnerabilities,web
npx @topy-ai/cso-skills list
Claude Code plugins: /plugin marketplace add TOPY-AI-LTD/TOPY-AI-CSO-Skills, then /plugin install topy-ai-cso-all@topy-ai-cso.
Codex: codex plugin marketplace add TOPY-AI-LTD/TOPY-AI-CSO-Skills. Pi: pi install npm:@topy-ai/cso-skills.
See the installer README.
Connect TOPY.AI
https://topy.ai/mcp/cso — OAuth with your TOPY.AI account (the same one as the AI CEO, CFO and COO). Only the accounts
that own the OneMan operation see its findings.
Client
How
Claude, ChatGPT (web, desktop)
custom connector https://topy.ai/mcp/cso
Claude Code
claude mcp add --transport http topy-cso https://topy.ai/mcp/cso
npx -y @topy-ai/cso-mcp after npx @topy-ai/cso-mcp login
Terminal and scripts
npm i -g @topy-ai/cso-cli && cso-cli login
Server checks (host-quick, host-audit) and passive checks of sites you don't control (public-check) are not on
topy.ai: they need a person on OneMan, with host consent granted in a terminal.
How the skills fit together
topy-cso-connect ─▶ topy-cso-setup (inventory · domain verify · consent) ─▶ topy-cso-goals
│
daily 06:17: topy-cso-daily-watch ─▶ topy-cso-cve-response ─▶ topy-cso-incident ─▶ topy-cso-fix-plan
│ topy-cso-dependencies · topy-cso-end-of-life
weekly Mon 07:23: topy-cso-weekly-report ─▶ the founder (email) and the CEO
│
before code ships: topy-cso-secrets · topy-cso-code-scan · topy-cso-threat-model
web (verified domains): topy-cso-domain-verify ─▶ topy-cso-web-checks ─▶ topy-cso-web-scan
before you trust it: topy-cso-package-vet · topy-cso-vet-skill · topy-cso-mcp-audit
Try: "Anything new I should worry about?", "Are we affected by CVE-2026-…?", "Did a key leak?",
"Is this MCP server safe to install?", "Send the weekly security report"
plugins/
topy-ai-cso/ core skills + .mcp.json (Claude Code) and mcp.json (Codex)
topy-ai-cso-vulnerabilities/ topy-ai-cso-code/ topy-ai-cso-web/
topy-ai-cso-agents/ topy-ai-cso-response/
topy-ai-cso-all/ bundle that depends on all six
.claude-plugin/marketplace.json Claude Code marketplace
.agents/plugins/marketplace.json Codex marketplace
skills.json machine-readable catalogue
packages/skills/ the @topy-ai/cso-skills installer
The skills are generated from a source catalogue kept next to TOPY.AI's MCP server; the build checks every tool a
skill names against the CSO server's tool list. Please open an issue rather than editing a SKILL.md.
Every morning: new CVEs, advisories and end-of-life dates for everything the company runs, matched to the stack — quiet unless something new needs a person.
The AI CSO's standing goals and its status report for the CEO: open critical exposure, secrets in code, verified-only checks, vetted skills and MCP servers, incident timelines.
First run for a new company or site: inventory the stack, verify each domain you own, and explain host consent — so every later check is allowed and complete.
Grade a verified site's HTTP security headers and cookies (A+ to F) and its TLS (protocols, certificate trust and expiry), with the exact fix for each.
OWASP MCP checks on the company's own MCP tool registry: read-only hints, destructive tools behind confirmation, scopes, audit logging and hosts it may call.
Before a skill, plugin or MCP server is installed or published: prompt-injection, data exfiltration, over-broad permissions and risky tools — the App Factory security gate.
Turn the findings into a reviewed runbook — each action tagged T1 (batch-safe) to T4 (stage first) — that a person reviews and runs. The CSO never applies it.
Record a security incident or CVE response from start to close — what happened, every step and who did it — so nothing is lost and the report writes itself.
Every Monday: what is open by priority, what was fixed, what changed in the stack, and at most three actions — for the Founder by email and for the CEO.
Henüz araç listesi yok. Uzak sunucular ilk sağlık kontrolünden sonra bir liste alır.
SSS
Sorular
TOPY.AI · AI CSO nasıl kurulur?
Bu sayfadan Cursor, Claude Desktop, Claude Code veya VS Code için http yapılandırmasını kopyalayın. PlugMCP proxy'sini değil, yayıncının kurulumunu kullanır.
TOPY.AI · AI CSO için API anahtarı gerekiyor mu?
Yayıncı şu ortam değişkenlerini listeliyor: TOPY_API_KEY.
TOPY.AI · AI CSO hangi aktarımı kullanıyor?
TOPY.AI · AI CSO sunucusuna HTTP üzerinden ulaşılır.
TOPY.AI · AI CSO en son ne zaman kontrol edildi?
Son kontrol: 2026-11-07. Founders' AI CSO: vulnerabilities, dependencies, secrets, verified-site checks, threat model, fixes.