fix(e2e): restrict readiness polling to loopback destinations

ARCHITECTURE.md

@@ -126,6 +126,9 @@ sequenceDiagram
   invents a hypothesis, observed result, or verdict.
 - Sandbox helpers copy the workspace, drop secret environment values unless
   explicitly allowlisted by **name**, and run subprocesses with `shell=False`.
+  Web E2E readiness URLs are loopback-only; see
+  [`docs/doctoring/sandboxed-web-readiness-loopback-boundary.md`](docs/doctoring/sandboxed-web-readiness-loopback-boundary.md)
+  and [`docs/adr/0003-sandboxed-web-readiness-loopback-boundary.md`](docs/adr/0003-sandboxed-web-readiness-loopback-boundary.md).
 - Logs and review receipts redact credential shapes (tokens, bearer values,
   known provider prefixes). They do not mask operational PII that the
   control plane must process.
@@ -173,3 +176,5 @@ resolver conflict.
   — product-specific psychometric repair heartbeat and scientific gates.
 - [`docs/doctoring/exact-artifact-sbom-attestation.md`](docs/doctoring/exact-artifact-sbom-attestation.md)
   — current increment's attestation decision and APA 7th citations.
+- [`docs/doctoring/sandboxed-web-readiness-loopback-boundary.md`](docs/doctoring/sandboxed-web-readiness-loopback-boundary.md)
+  — loopback-only web E2E readiness polling and APA 7th citations.

CHANGELOG.md

@@ -84,6 +84,12 @@ Semantic Versioning where the repository publishes a release.
 
 ### Fixed
 
+- Web verification now checks services through local readiness addresses only.
+  Start the backend and frontend on this computer and use their local health
+  URLs when running the check.
+- Review results now separate cosmetic notices from blocking failures. Open the
+  failure details and correct the requested issue before running the check
+  again.
 - Resolve Strix visibility from the trusted GitHub event for ordinary push,
   schedule, and pull-request runs, reserving API retries for cross-repository
   dispatches whose workflow token may not see the target repository.

docs/adr/0003-sandboxed-web-readiness-loopback-boundary.md

@@ -0,0 +1,9 @@
+# ADR-0003: Sandboxed web readiness loopback boundary
+
+- Status: accepted
+- Date: 2026-08-25
+- Scope: ContextualWisdomLab/.github control-plane E2E sandbox
+- Decision: Poll `--backend-ready-url` and `--frontend-ready-url` only after the URL is proven to be HTTP(S) loopback. Accept literal `localhost` or a standard-library loopback address, resolve `localhost` and require every answer to be loopback, unwrap IPv4-mapped IPv6, reject userinfo and missing hosts, and keep redirects disabled.
+- Ownership: `.github` owns the sandbox helper. Product repositories keep pointing readiness at their own loopback services.
+- Figma File ID: N/A. This repository has no customer UI.
+- Consequence: A review run cannot use the sandbox poller as an SSRF trampoline to metadata services or public hosts. Operators fix a rejected URL by pointing it at `127.0.0.1` or `::1`. Papers live in `docs/doctoring/sandboxed-web-readiness-loopback-boundary.md`.

docs/doctoring/sandboxed-web-readiness-loopback-boundary.md

@@ -0,0 +1,85 @@
+# Sandboxed web readiness loopback boundary
+
+## Decision
+
+`sandboxed_web_e2e.py` opens a readiness URL only after
+`require_loopback_readiness_url` accepts it. The accepted destinations are
+literal `localhost` (a trailing FQDN dot is stripped) or an address that
+Python's standard-library `ipaddress` module classifies as loopback after
+IPv4-mapped IPv6 addresses are unwrapped. Literal `localhost` is then
+resolved; every A/AAAA answer must itself be loopback, so a poisoned hosts
+file cannot smuggle a public address through the name allowlist. Redirects
+remain disabled.
+
+This supports the complete IPv4 loopback block, including `127.0.0.2`, and
+IPv6 `::1`. It rejects `0.0.0.0`, `::`, public hosts, `.localhost`
+subdomains, cloud-metadata link-local addresses, missing hosts, and
+userinfo-confused URLs such as `http://user@127.0.0.1/`. A mapped public
+address such as `::ffff:8.8.8.8` cannot pass merely because it is IPv6.
+
+The boundary uses the standard library rather than a second address table.
+It therefore follows the runtime's maintained special-purpose definitions and
+keeps one fail-closed validation point before any network request. Do not add
+individual non-loopback exceptions.
+
+This successor lands the same buyer-facing repair as ContextualWisdomLab/.github#1244
+on current `main` and keeps Strix classifier ownership out of the SSRF slice
+(unlike ContextualWisdomLab/.github#1313).
+
+## Operator next action
+
+Point `--backend-ready-url` and `--frontend-ready-url` at the sandboxed
+service on loopback. If readiness fails with `URL cannot target external
+hostname`, replace the destination with `http://127.0.0.1:<port>/...` or
+`http://[::1]:<port>/...` instead of opening the firewall or adding a
+hostname exception.
+
+## Verification
+
+The regression exercises literal `localhost`, a trailing-dot `localhost.`,
+`127.0.0.1`, another address in `127.0.0.0/8`, IPv6 `::1`, mapped loopback
+`::ffff:127.0.0.1`, an unspecified address, a `.localhost` subdomain, a
+public hostname, the common cloud metadata address, mapped public IPv6,
+userinfo, a missing host, and poisoned localhost resolution (public A,
+mapped public AAAA, empty answers, resolver errors, and non-IP answers).
+The existing no-redirect test continues to prove that an allowed readiness
+endpoint cannot redirect the poller across the boundary.
+
+```mermaid
+flowchart TD
+  Url["Readiness URL"]
+  Scheme{"http or https?"}
+  Userinfo{"userinfo present?"}
+  Host{"loopback IP, or localhost whose every resolved answer is loopback?"}
+  Open["Poll with redirects disabled"]
+  Reject["Fail closed before any request"]
+
+  Url --> Scheme
+  Scheme -->|"no"| Reject
+  Scheme -->|"yes"| Userinfo
+  Userinfo -->|"yes"| Reject
+  Userinfo -->|"no"| Host
+  Host -->|"no"| Reject
+  Host -->|"yes"| Open
+```
+
+## References
+
+Berners-Lee, T., Fielding, R., & Masinter, L. (2005). *Uniform Resource
+Identifier (URI): Generic syntax* (RFC 3986). Internet Engineering Task
+Force. https://doi.org/10.17487/RFC3986
+
+Cotton, B., Vegoda, L., Bonica, R., & Haberman, B. (2013). *Special-purpose
+IP address registries* (RFC 6890). Internet Engineering Task Force.
+https://doi.org/10.17487/RFC6890
+
+Internet Assigned Numbers Authority. (2026). *IANA IPv4 special-purpose
+address registry*. Retrieved August 25, 2026, from
+https://www.iana.org/assignments/iana-ipv4-special-registry/iana-ipv4-special-registry.xhtml
+
+OWASP Foundation. (n.d.). *Server-side request forgery prevention cheat
+sheet*. Retrieved August 25, 2026, from
+https://cheatsheetseries.owasp.org/cheatsheets/Server_Side_Request_Forgery_Prevention_Cheat_Sheet.html
+
+Python Software Foundation. (2026). *ipaddress — IPv4/IPv6 manipulation
+library*. https://docs.python.org/3/library/ipaddress.html

scripts/ci/sandboxed_web_e2e.py

@@ -3,16 +3,19 @@
 from __future__ import annotations
 
 import argparse
+import ipaddress
 import json
 import os
 import signal
 import shutil
 import shlex
+import socket
 import subprocess
 import sys
 import tempfile
 import time
 import urllib.error
+import urllib.parse
 import urllib.request
 from collections.abc import Sequence
 from dataclasses import dataclass
@@ -115,12 +118,61 @@ def start_service(label: str, command: str, cwd: Path, env: dict[str, str], logs
     return Service(label=label, command=command, process=process, log_path=log_path)
 
 
+def _require_loopback_ip_text(ip_text: str, hostname: str) -> None:
+    """Reject a literal or resolved address that is not loopback."""
+    try:
+        address = ipaddress.ip_address(ip_text)
+    except ValueError as exc:
+        raise ValueError(f"URL cannot target external hostname: {hostname}") from exc
+    if address.version == 6 and address.ipv4_mapped is not None:
+        address = address.ipv4_mapped
+    if not address.is_loopback:
+        raise ValueError(f"URL cannot target external hostname: {hostname}")
+
+
+def _require_resolved_loopback_hostname(hostname: str) -> None:
+    """Resolve a literal localhost name and require every answer to be loopback."""
+    try:
+        results = socket.getaddrinfo(hostname, None)
+    except OSError as exc:
+        raise ValueError(f"URL cannot target unresolved hostname: {hostname}") from exc
+    if not results:
+        raise ValueError(f"URL cannot target unresolved hostname: {hostname}")
+    for result in results:
+        _require_loopback_ip_text(result[4][0], hostname)
+
+
+def require_loopback_readiness_url(url: str) -> None:
+    """Reject a readiness URL that is not a local loopback HTTP(S) target.
+
+    Operators should point ``--backend-ready-url`` and ``--frontend-ready-url``
+    at the sandboxed service itself. Public hosts, cloud metadata addresses,
+    unspecified bind addresses, DNS names other than literal ``localhost``,
+    and userinfo-confused URLs are rejected before any request is opened.
+    Literal ``localhost`` is resolved and every answer must be loopback, so a
+    poisoned hosts file cannot smuggle a public A/AAAA record through the
+    name allowlist. IPv4-mapped IPv6 addresses are unwrapped and re-checked
+    so ``::ffff:8.8.8.8`` cannot bypass the loopback rule.
+    """
+    parsed = urllib.parse.urlparse(url)
+    if parsed.scheme.lower() not in {"http", "https"}:
+        raise ValueError(f"URL must start with http:// or https://, got: {url}")
+    if parsed.username or parsed.password:
+        raise ValueError("URL cannot include userinfo")
+    hostname = (parsed.hostname or "").lower().rstrip(".")
+    if not hostname:
+        raise ValueError("URL must include a loopback hostname")
+    if hostname == "localhost":
+        _require_resolved_loopback_hostname(hostname)
+        return
+    _require_loopback_ip_text(hostname, hostname)
+
+
 def wait_for_url(url: str, timeout: int, service: Service) -> bool:
     """Poll a readiness URL until it responds or the service exits."""
     if not url:
         return True
-    if not (url.startswith("http://") or url.startswith("https://")):
-        raise ValueError(f"URL must start with http:// or https://, got: {url}")
+    require_loopback_readiness_url(url)
     deadline = time.monotonic() + timeout
     opener = urllib.request.build_opener(NoRedirectHandler())
     while time.monotonic() < deadline: