Skip to content

CHARTER.md

Latest commit

 

History

History
110 lines (93 loc) · 5.58 KB

CHARTER.md

File metadata and controls

110 lines (93 loc) · 5.58 KB
Raw

CHARTER — browser

  • Status: ACTIVE (product reframed 2026-07-14)
  • Scope confirmed: 2026-07-10
  • Updated: 2026-07-14
  • Repository: marcohost33-maker/browser

PRODUCT REFRAME (2026-07-14) — read first. Marco's decision of 2026-07-14 resets the north star: browser is a native, offline-capable browser/webapp runtime program that executes arbitrary foreign web applications locally, replacing cloud hosting, and running without any AI layer. browser is standalone (ADR-008, 2026-07-16): nigin-engine and browser-nigin are separate independent repositories linked only by knowledge transfer, not an architectural stack and not dependencies. Capability is delivered in staged trust classes — T1 (owner-controlled packages) → T2 (curated third-party) → T3 (arbitrary foreign web content); the accepted north star is T3. The binding reframe record is ADR-005 (offline runtime trust classes), ADR-006 (runtime evaluation protocol) and ADR-007 (signed package evaluation); see PR #22 — not duplicated here. Runtime-direction consideration for T3: a Chromium-based engine (CEF/Electron) is favoured for inherited site-isolation and a maintained engine security-patch path over a Tauri/system-WebView approach; the engine security-patch SLA is a central cut criterion. The binding runtime-framework choice is deferred to the measured comparison in ADR-006 — no framework is accepted here. Everything below that describes browser as a "public MCP client webapp consuming remote endpoints" is superseded product framing, retained because its security/privacy/governance substance stays valid; it is overruled by ADR-005/006/007 where the two conflict.

Purpose

browser is a native, offline-capable runtime that executes locally installed web applications, staged from owner-controlled (T1) toward arbitrary foreign web content (T3). It replaces cloud/edge hosting for those apps and runs without an AI layer. It is not browser-nigin (the AI layer) and does not define or own the contract core (nigin-engine).

Historical framing (superseded 2026-07-14): "browser is the public, privacy-first browser application and MCP client of the Vero/Nigin system, consuming a versioned MCP contract supplied by ENG-01." Retained for provenance.

Product principles

  • Explicit user control. Connections, capabilities and requests remain visible, bounded, revocable and attributable to a user action.
  • Privacy by default. Minimize data, retention, permissions, third-party dependencies and network destinations.
  • Fail closed. Unsupported security-critical semantics, unapproved origins, invalid contracts and consent drift block execution.
  • Standards-first web application. Use browser-native capabilities where practical; select a framework only with measured evidence.
  • Low operating cost. The default product should avoid unnecessary paid APIs and infrastructure, without weakening security or reliability.
  • MCP is internal and optional (ADR-008). MCP consumption belongs in browser as an internal, optional capability, off the T1 critical path. Any signed contract it consumes may be sourced internally or from any signed producer — no external nigin-engine dependency is assumed. Signature/provenance-verification discipline is retained and re-pointed at "whatever signed contract browser consumes".
  • Evidence before claims. Designed, implemented, verified and production-ready are separate states.
  • Open-source candidate. Public release occurs only after security, privacy, legal, accessibility, operational and repository-governance gates pass.

M1 objective

Validate one primary user and one read-only task, then deliver the smallest browser slice that can:

  1. use an endpoint class approved by ADR-003;
  2. identify the endpoint, origin, protocol revision and trust tier;
  3. negotiate and display capabilities as untrusted data;
  4. obtain explicit approval for one bounded read-only request;
  5. enforce timeout, cancellation, payload and rendering limits;
  6. return a validated result or normalized error;
  7. clear all sensitive session state.

Architectural boundaries

  • Presentation cannot call a transport directly.
  • Endpoint, credential, capability, consent and storage policy are independent enforcement boundaries.
  • CSP origins are deployment policy; a user-entered endpoint cannot widen them.
  • Runtime code cannot proceed until product, endpoint/deployment and contract gates are satisfied.
  • Remote production traffic requires HTTPS. Plain HTTP is limited to explicit loopback development endpoints.

Non-goals for M1

  • defining an MCP protocol or ENG-01 contract;
  • Wasmtime, WIT, WASI, CAS, solver, orchestrator or engine fabric;
  • an unrestricted universal MCP browser or marketplace;
  • arbitrary user-supplied remote endpoints;
  • write tools, autonomous execution, sampling or elicitation;
  • persistent credentials or opaque background synchronization;
  • a PWA/service worker before its cache and sensitive-data threat model passes.

Source-of-truth map

  • Execution roadmap: docs/ROADMAP.md
  • Program roadmap: docs/MASTER_ROADMAP.md
  • Current decisions: docs/OPEN_DECISIONS.md
  • Architecture decisions: docs/adr/
  • Security profile: docs/security/
  • Current implementation state: docs/IMPLEMENTATION_STATUS.md
  • Vulnerability reporting: SECURITY.md

The historical cwzl document that described Wasmtime/WIT/CAS/Engine-Fabric work is retained only as provenance in docs/007_ANALYSE_platform-track_PROVENIENZ.md; its implementation owner is ENG-01, not APP-01.