8000 Website: security page documenting the guard's threat model by tashdroid · Pull Request #1 · tashfeenahmed/cashmaxx · GitHub
[go: up one dir, main page]

Skip to content

Website: security page documenting the guard's threat model 8000 - #1

Open
tashdroid wants to merge 1 commit into
tashfeenahmed:mainfrom
tashdroid:feat/site-security-threat-model
Open

tashdroid wants to merge 1 commit into
tashfeenahmed:mainfrom
tashdroid:feat/site-security-threat-model

Conversation

@tashdroid
Copy link
Copy Markdown

What

Adds a /security/ page to the Cashmaxx website (cashmaxx.neu.so) stating the guard's threat model, and links it from every page footer, the landing page's Safety section, and sitemap.xml.

Why

Serious self-hosting buyers compare against custodial pitches (e.g. MetaMask Agent's advertised $10k transaction protection + 2FA, Coinbase's KYT screening + TEE key isolation, June 2026). The landing page lists "budget caps · approvals · kill switch" but never explains the actual boundary — that the agent never sees wallet, payment, email or social credentials. This page answers that question directly, and honestly positions self-hosted key separation against insured custody instead of pretending to compete with it.

Contents (all grounded in the code, not invented)

  • Key separation: guard-only secrets in ~/.cashmaxx/guard.json (0600); agent holds only an agent token (SHA-256 digest, constant-time compare; no approve/patch/unfreeze scope); owner PIN as scrypt hash, in-memory 12h sessions; guard Telegram bot on its own token, owner chat id checked. — cashmaxx/guard/auth.py, docs/cashmaxx/ARCHITECTURE.md
  • What triggers approval: the 9-step fixed-order policy (frozen / invalid / unsupported kind / budget / daily cap / hourly rate → deny; over threshold / new recipient → needs approval), re-check of hard limits at approval execution, 24h approval expiry, fail-closed when the guard is down. — cashmaxx/guard/policy.py, watchers.py
  • What the freeze does: first check in the policy engine; pullable by anyone (WebUI, Telegram, CLI, agent tool); owner-scope-only unfreeze; automatic loss stop on 7-day net below threshold. — app.py, watchers.py
  • "What this is not": states the real limits (guard protects spending not the OS; run guard as separate user/container; OS-level compromise breaks everything; chain payments irreversible; guard binds 127.0.0.1).

How it was tested

  • Static site, no build step. All site HTML parsed with a strict tag-balance checker (python html.parser): passes.
  • Programmatic internal-link check across every page under site/public: no broken links; new /security/ URL resolves via site/public/security/index.html.
  • Content cross-checked line by line against cashmaxx/guard/policy.py, auth.py, app.py, watchers.py, config.py and docs/cashmaxx/ARCHITECTURE.md; no claim on the page that the code doesn't enforce.

Board card #828

Adds /security/ to the site: key separation (the agent holds only an
agent token; wallet, payment, email and social credentials live only in
the guard), what the agent can and can never do, the fixed-order policy
that triggers approvals, and exactly what the freeze and loss stop do.
Links it from every footer, the safety section, and the sitemap.

Board card #828
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant

0