gstack/docs/gstack-2/STATUS.md

GStack 2 status

Status at the 2026-07-17 documentation checkpoint: BLOCKED.

The candidate contains substantial implementation, but it is not a released or verified GStack 2. DONE is prohibited until every P0 gate is backed by the required evidence layer. Current blockers include a CoreDevice-compatible signed physical iPhone, native Windows CI, a passing live v3 installed-host adversarial run, native Linux evidence, and the signed physical-iPhone loop. No branch push, draft PR, or PR-ready claim is authorized by this status.

Implemented candidate surface

• Pinned audit base: bb57306d98c97011b0919c6132705a15b1579781.
• Baseline counts and pre-existing failures captured without relabeling them candidate regressions.
• 755 unique open issue/PR records reconciled and deterministically mapped; all 16 required upstream PR snapshots traced.
• Exactly six canonical skill directories: plan, design, qa, debug, review, and ship.
• The six /plan top-level modes are exactly Discovery, Product, Engineering, DX, Specification, and Full chain.
• Standard installer matrix is green: 470/470 checks, 16 installs and two removals with CLI 1.5.19. Project/global installs passed for Claude Code, Codex, Cursor, Pi, OpenClaw, and GitHub Copilot; selected-skill and opt-in compatibility-alias cases, paths with spaces, source symlink, physical copies, and canonical hashes passed. The committed artifact is evals/installation/install-matrix.json.
• Current six names/descriptions measure 982 characters (about 246 four-character token-equivalents), roughly 77.6% below the correctly parsed baseline of about 1,100 in the regenerated tree.
• Generated inventory contains 55 preserved modules, 16 carved sections, 25 scenarios, 16 regression definitions, and 78 assets.
• Compatibility aliases remain opt-in and outside default six-skill discovery; each prints its replacement and contains no copied judgment.
• Judgment provenance, behavioral contracts, 25 structured scenarios, and 16 upstream bug-fix regression definitions implemented.
• The 2026-07-17 regenerated parity rerun is green: 4,681 checks covering 55 modules, 16 sections, 25 scenarios, 16 regressions, and 78 assets.
• Deterministic semantic parity is green: 295 checks across 14 suites, 15 executions, 15 comparison dimensions, 16 carved sections, and nine authority-policy unit cases, including unsupported numeric claims. Exact preserved source bodies are the primary oracle. These are deterministic policy checks, not behavioral-adversarial proof, and do not close the installed-host gate. All three retained Claude Haiku live samples are classified REGRESSION; they are preserved as noisy supplemental evidence, never cherry-picked as a primary gate or represented as green.
• The focused macOS GStack 2 suite is green: 136 pass / 0 fail, 1,128 assertions across 15 files. Log: /tmp/gstack2-test-command-candidate-final2.log. This focused surface does not substitute for the broad or native-platform gates.
• Optional host-neutral runtime implemented with canonical paths, repo/worktree state identity, locks, atomic writes, effect claims, doctor/config/state/cleanup, migrations, upgrade/rollback, and uninstall.
• Managed runtime installer coverage is green at 24 pass / 0 fail and 336 assertions. The deterministic clean macOS arm64 managed-bundle audit records 110 components, 1,829 files, 450,044,315 bytes, and 50 capability launchers. This is a platform-specific bundle measurement, not a universal byte count; platform-native package payloads differ. Setup installs frozen production-only dependencies; the development-only Claude Agent SDK is excluded. The Sharp/ngrok closure is included. The Hugging Face sidecar is excluded and its package is development-only, so setup installs neither its inference runtime nor model weights and reports the L4 capability unavailable. A clean Linux arm64 container smoke also used the production-only install with the development SDK absent, completed a local browser journey and Sharp full-page screenshot, and uninstalled while preserving state.
• Filesystem lifecycle coverage passes for clean install/uninstall, paths with spaces, source symlinks with internal-link rejection, read-only destination reporting on macOS, interrupted-pointer rollback, crash-journal repair, and last-known-good launcher recovery.
• Crash/resume external-effect idempotency is covered through an actual local Git push: the effect executes at most once, and resume refuses to repeat a command that may already have happened.
• Context.dev public-URL/consent/failure contract is green at 22 pass / 0 fail and 139 assertions; deprecated search is typed unsupported rather than fabricated. context options and explicit context select host|local-browser|none choices persist without granting Context.dev consent. A verified-key live smoke passed the official Markdown scrape endpoint using protected input and an isolated temporary home; the key was not persisted in the repository or permanent GStack state.
• iOS candidate fixes implemented for UDID/CoreDevice identity, malformed device-list errors, suspended-app bounded timeout, and active-bundle checks.
• Default/free-test roots now include design/test and ios-qa/daemon/test.
• Focused retained capability suites are green: iOS daemon 95 pass / 0 fail / 229 assertions; design 101 pass / 0 fail / 381 assertions; PDF 189 pass / 0 fail / 398 assertions; and diagram 51 pass / 0 fail / 1 skip / 120 assertions. The opt-in paid diagram lane recorded two skips and is not live provider evidence.
• Physical-iOS preflight recorded 9 pass / 0 fail / 1 deploy skip and 29 assertions. The direct smoke then returned typed code signing_unavailable, category setup_gate; it installed no app and wrote no pass artifact. A later attempt against the explicitly selected iPhone10,6 on iOS 16.7.10 returned device_not_wired; legacy USB pairing validates, but CoreDevice pairing is unsupported. The other connected phone was not used.
• Installed-host adversarial evidence is retained without relabeling: v1 failed; immutable v2 failed even though QA passed because the classifier produced false negatives for debug, review, and ship; v3 offline harness coverage is green at 18 pass / 0 fail and 111 assertions. Live v3 has not run and has no artifact, so there is no passing live installed-host adversarial result.
• A current managed-bundle audit confirms no model-weight download path in setup. The standard Agent Skills installation remains Markdown-only and independent of the optional runtime.
• An isolated local-browser journey completed navigation, snapshot, screenshot, status, and stop cleanup. The stop acknowledgement regression has a 2 pass / 0 fail focused test.
• The uninterrupted macOS broad suite is green under singleton isolation: 6,240 pass / 226 expected skips / 0 fail and 25,449 assertions across all 383 files. This includes the complete local-browser suite. The local Windows-safe singleton lane is also green at 2,815 pass / 57 expected skips / 0 fail and 8,586 assertions across all 213 curated files; it is not native Windows evidence.
• Architecture, privacy, Context.dev, host compatibility, upgrade/rollback, migration, and governance documentation added.

Blocking or incomplete P0 evidence

• Record a passing live v3 installed-host adversarial run and execute the installed skill in each representative host UI. The six hosts are Verified at the installer layer only; v1 and v2 are retained failed runs, and v3 is offline-harness evidence only.
• Prove runtime-absent judgment behavior through an actual host invocation. Runtime install failure and the real default capability lifecycle are covered.
• Complete native Linux and Windows matrices. The uninterrupted macOS broad singleton run is green at 6,240/226/0 across 383 files, and the local Windows-safe singleton lane is green at 2,815/57/0 across 213 files. Neither substitutes for native execution on its named platform. The candidate workflow targets GitHub-hosted ubuntu-24.04 and windows-latest, but remote execution requires explicit authorization for a CI-only ref because this candidate is intentionally local-only before P0 passes.
• Integrate the passing complete local-browser suite and live journey into final cancellation/leak evidence. The production dependency and managed bundle audit contains no cloud-browser provider/path.
• Complete the physical-iPhone five-check loop with a signed test app. The unsigned Release guard and earlier preflight passed, but the latest authorized target cannot expose the required CoreDevice service on iOS 16.7.10. Provide a CoreDevice-compatible iPhone without broadening the locked backend. No app was installed and no pass artifact exists.
• Verify in one real cancellation run that no subprocess, listener, credential, browser, or device-session remains after interruption. Existing component tests cover each cleanup contract separately but not this aggregate live gate.
• Finish final evidence-linked disposition for every infrastructure item; see the 25-row table in ARCHITECTURE.md. Current focused evidence does not replace the remaining live and native-platform gates.

Evidence index

Evidence	Path	State
Measured baseline	BASELINE.md	Recorded
Candidate and baseline command ledger	TEST-EVIDENCE.md	Broad macOS and local Windows-safe singleton runs green; live/native-platform gates pending
Complete skill migration	SKILL-MIGRATION.md	Generated; 55/55 assignments
Judgment provenance	JUDGMENT-PROVENANCE.json	Generated; 4,681-check parity rerun green
Parity contract	JUDGMENT-PARITY.md	Green for source/render/contract/asset fixtures
Semantic parity	SEMANTIC-PARITY.md	Deterministic 295-check corpus green; retained live samples are regressions
Installed-host adversarial	eval overview, immutable v2 artifact	V1 failed; immutable V2 failed; V3 offline 18/111 green, live V3 not run; no passing live run
Structured scenarios	SCENARIOS.md	25/25 structured routing fixtures green
Backlog traceability	BACKLOG-MAP.json	755 unique items mapped
Context integration	CONTEXT-DEV.md	Automated contract 22/139 green; verified-key official-endpoint live smoke passed
Host matrix	HOST-COMPATIBILITY.md	470/470 checks; six hosts Verified at installer layer; live v3/UI launch pending
Privacy boundary	PRIVACY.md	Implemented contract; full retained-tool egress audit pending
Physical iOS	IOS-PHYSICAL-DEVICE.md	Earlier preflight 9 pass / 1 deploy skip / 29 assertions; latest selected target lacks a supported CoreDevice service; no app/pass artifact
Upgrade/recovery	UPGRADE-AND-ROLLBACK.md	Runtime installer 24 pass / 336 assertions; deterministic clean macOS arm64 bundle audit recorded

Interpretation rules

• MECHANICAL_PORT means the pinned rendered judgment body remains equal after normalization; it is not permission to rewrite prose.
• BUG_FIX requires its linked PR/reproduction and regression fixture.
• NEEDS_EVIDENCE, DEFER_COMMUNITY, and SUPERSEDED_BY_CONSOLIDATION are auditable dispositions, not GitHub state changes. No labels, issues, or PRs are mutated by the map generator.
• A fixture-backed structural result does not replace a live browser, physical device, external account, native OS, or host-install result where the gate explicitly requires one.
• Do not market or release this branch as GStack 2 while this status is BLOCKED.