SpaceXSI refresh validation — 2026-11-05

Scope: the historical spacexai family at base 99eaa3d1a38ce28ed421ac4020e37e898ea98c9e. Current naming and product-plane decisions are in the source reconciliation, with a separate Build reconciliation.

Tier 0+1 only. No tenants declared; VERIFIED-LIVE = 0. Documentation reads, synthetic transports, local parsing and provider schema validation do not establish tenant enforcement. All seven guides remain ai-drafted. No product sign-in, setting, credential, integration, Terraform plan/apply/import or real agent execution occurred.

Coverage

Guide	Leveled controls	Include tags	Numbered pack files	Static rendered pack groups / code excerpts
Cursor / Common Controls	34	24	24	27 / 38
Grok Bot	27	17	30	28 / 44
Grok workspace	9	0	0	0 / 0
Grok API	11	4	4	4 / 4
Grok Build	9	8	9	9 / 9
Cursor Cloud Agents	10	2	2	2 / 2
Cursor Origin	9	5	7	7 / 7
Total	109	60	76	77 / 104

Baseline: two guides, 48 controls, 53 numbered packs, 40 include tags. Result: five new guides, 61 additional controls, 23 additional packs and 20 additional include tags. Helpers, READMEs and Terraform provider/variable files are excluded from numbered-pack totals. A rendered include can contain several pack types and several marked regions; Cursor also retains its existing GitHub pack include. These measures are not interchangeable.

Completed checks

• bun test ./scripts/test-spacexai-refresh-packs.ts ./scripts/test-grok-build-packs.ts: 27 passed, 0 failed, 364 assertions. Tests cover transport/redirect/target restrictions, response redaction, typed schema drift and pagination, authoritative team identity, partial evidence returning UNKNOWN, native TOML parsing and policy-input injection, opt-in native read commands and private partial output. New Bash wrappers are actually executed without credentials, proving they reach the helper’s structured UNKNOWN precondition before network access. Synthetic fixtures do not prove a live API response. Additional private curl-fixture tests verify the Bot helper’s fixed HTTPS origin, explicit synthetic-only loopback escape, rejected unsafe credential/path representations, first-argument -q, request/size limits, no redirect forwarding and suppressed error bodies; they inspect invocation semantics rather than asserting a live TLS exchange.
• bash scripts/validate-guides.sh: all tests passed, 0 warnings. Additional local source checks found canonical profile labels, nonempty Description and Rationale sections, and no fenced code in all seven guides. Existing control numbers and anchors remain intact.
• bash scripts/sync-packs-to-data.sh: completed. Unrelated files were restored only after confirming their sole difference was the generated timestamp. Full pack validation confirmed fresh generated YAML, reachable marked packs, no unresolved pack anchors or dead includes, and no declared-mode contradictions.
• bash scripts/validate-packs.sh --all: 5 existing failure groups, 6 warnings, with no failure attributed to this lane. Existing failures are three section/type collisions (Bitbucket, ChatGPT Enterprise, HashiCorp Vault), two Bash syntax files (HashiCorp Vault and Docker Hub), 11 extension/type violations in other families, and one prose-only Azure DevOps Terraform file. The post-commit --touched gate passed on 66 relevant changed files, with no failures in this branch’s files.
• Native cursor/cursor 0.8.0 provider initialization and terraform validate passed for both Origin and Cloud Agents, using Terraform 1.17.1 and an isolated empty CLI configuration with vendor credential variables absent. Provider installation reported developer signing key 5E389DAC0B402A90. Formatting passed. No provider configure, plan, apply or import ran. This verifies the provider’s schema, not actual tenant authorization, policy effects or recovery access.
• Local Jekyll build passed using the existing cached Ruby 3.2 bundle and a separate validation Gemfile. The repository’s Gemfile/lockfile were unchanged; CI’s locked dependency installation remains separate evidence.
• Static parsing of the built HTML found the expected 109 full-guide profile records, 77 rendered pack groups and 104 nonempty code excerpts, with no broken guide/fragment links from the seven guides. This is static HTML evidence only.
• OCEAN build --validate passed 197 checks against the existing canonical checks repository. No OCEAN check or schema file changed in this lane. The staged source diff passes whitespace checks when generated pack YAML is excluded. The generator emits indented empty lines and a final blank line in new YAML files, so an unrestricted staged git diff --check reports those canonical generated lines. They were preserved to maintain exact generator freshness; no global generator behavior changed.
• The dedicated SpaceXSI Pack Safety workflow runs the two explicit Bun suites for changes to their test files, these pack directories or the workflow. A two-product matrix runs Terraform 1.17.1 format/init/validate with the native cursor/cursor 0.8.0 constraint, empty Terraform CLI configuration and no vendor credentials. It runs no plan/apply/import. Permissions are contents: read, checkout credentials are not persisted, and each job has a ten-minute bound. Local actionlint passes; both Bun suites still report 27 passed / 364 assertions. The existing shared Test workflow did not run these suites; it remains unchanged. At signed commit 1e32d88135f0089e8ac78fd352dc080150510c3a, CI run 42085004025 passed all three jobs. Actual Linux job logs contain 27 passed / 0 failed / 364 assertions, both pinned provider installations with developer key 5E389DAC0B402A90, and both “Success! The configuration is valid.” results. Shared Test and Semgrep also passed at that commit. Later PR-head results are reported separately in the handoff.

Browser evidence and remaining checks

The local preview served successfully on loopback. Interceptor’s first tab_create timed out; a subsequent group read confirmed no attached tabs. A second tab_create also timed out without a successful attached-tab result. No shared daemon or browser settings were changed. The CUA fallback reported browsers: [], and creating a hidden IAB tab returned “Browser is not available: iab”.

An independent Factory-lane agent subsequently reused its own browser tab/group for a fresh preview on 127.0.0.1:63665, serving /tmp/hth-spacexai-render-a66a460 built from signed commit a66a46070ab66a99667cbe47ba5ff0759834c72d. A Git diff verified unchanged guide/pack file blobs through the CI/evidence follow-up; the earlier preview and its failures are not final-source proof. The bounded fresh-page results are:

Guide	Cheat rows / populated cells	Empty cells	Code excerpts	Full view / representative code expansion
Cursor	34 / 102	0	38	Correct full-view heading; one nonempty block expanded
Grok Bot	27 / 81	0	44	Correct full-view heading; two nonempty blocks expanded
Grok workspace	9 / 27	0	0	Full view read; no pack to expand
Grok API	11 / 33	0	4	Full view read; one nonempty block expanded
Grok Build	9 / 27	0	9	Full view read; one nonempty block expanded
Cursor Cloud Agents	10 / 30	0	2	Full view read; one nonempty block expanded
Cursor Origin	9 / 27	0	7	Full view read; one nonempty block expanded
Total	109 / 327	0	104	Representative expansion on all six guides with packs

Each native table had every cell populated; actual cell samples were read. Initial compact/full-text reads for Cursor and Bot timed out, then narrower fresh DOM queries established their headings, row counts, cell count/emptiness and expanded nonempty code. Those queries are not a claim that every cell was manually read. The peer artifacts are results.json, narrow-retries.json and page snapshots under the Factory worktree’s .hth-validation/factory-refresh/cross-lane-final/. Clipboard/download behavior, responsive layout and screenshots remain unverified; the browser pass did not exercise them. Static code/excerpt/link checks do not replace those checks. Existing user tabs and application focus were preserved; no shared daemon or browser settings were changed. This local-site browser work is still Tier 0+1 and does not establish any live vendor control.

Source and execution limits

Current first-party documents support separate Cursor, Bot, Grok workspace, API, Build, Cloud Agent and Origin authority planes. They do not establish inherited Cursor policy for a Grok API/workspace tenant. Full X social/account/developer and standalone X Chat administration remain uncovered external operational scope, with no existing X/Twitter guide in this base; current Grok FAQ distinguishes X operations from SpaceXSI. The AI integration’s developer-account creation, personal-only Bot linking, separate training setting and AI group-message exposure are covered. Grokipedia administration, negotiated government/partner deployment settings, SIP-specific hardening, hosted Build publication details and a completed Graphite acquisition remain explicit source gaps or external deployment boundaries. Some new native controls are beta, rollout-dependent or Enterprise-only. The provider’s registry web page/latest-release state was not established; exact-version native initialization and developer-signature verification did pass, alongside first-party GitHub source and installed-schema reads.

No pack’s metadata inventory, configuration candidate, read-only output or successful local test is whole-control compliance. Bot checks with an unobservable surface return UNKNOWN/exit 3. The destructive offboarding pack requires authoritative team-ID proof and retains the computer/disk-only deletion action; the newly documented broader Bot-data erase is described separately in the guide.