phoebus_os/docs/HANDOFF.md
Phoebus 3b219fc851
Some checks failed
ci / test (push) Has been cancelled
fix(agents): harden both user-level agent paths
2026-10-12 07:15:31 -07:00

90 KiB
Raw Blame History

Handoff

Rolling note between work sessions. Update this file before you finish, then commit it. It should be short and current; use git log for detail.

  • Both user-level paths (2026-10-12): See USER-AGENTS.md / ADR 0016. Native desktop configuration now works with durable conflict-aware undo; MCP stages actual reviewed package plans without elevation. Workspace agents have read-only tool grants and isolated networking by default. Personal history is private/concurrent and desktop context is separate; bridge handshake/reconnects are bounded and never replay pending work. Bundle both user binaries and Bubblewrap. Concurrent descriptor inheritance and lock-release races fixed. Full gate passes 991 Rust + 22 harness tests, strict Clippy, fmt, secret scan, dynamic/static builds, three UI handler checks and two real host isolation probes. Gentoo user fixture run serial-boot-20261012T141203Z-b2ac2870 passed 11/11 with clean poweroff. Next: fresh bundle installation, live visual reload/real-seat UI acceptance, actual-provider workflows and narrow host-held MCP/credential transport. Do not expose host home/runtime to the workspace jail. Base and ISO unchanged.

  • Agentic hardening (2026-10-12): All seven findings have fixes and regressions; see AGENTIC-HARDENING.md / ADR 0015. Native Gentoo run serial-boot-20261012T105530Z-9c09094e passed 14/14 with clean poweroff: developer/user/root isolation, descriptor snapshot, Portage exact install/ removal, Flatpak commit install/removal, Nix store-path verification and persistent kernel lock. Two earlier runs had fixture errors and are not product acceptance passes. A later command lock lease closes broker-death exclusion loss; its real-process regression passes. Final static smoke serial-boot-20261012T112454Z-47d3cfae passed 8/8, including network-free history/snapshot and repeated sandbox denials. Gates passed: 981 Rust tests, 22 harness safety tests, strict Clippy, fmt, secret scan, and dynamic/static release builds. QEMU base image and ISO unchanged. Initial space failure was resolved by generated compiler-cache cleanup. Next acceptance: fresh installation, native Snap revision/drift, interrupted native worker and broader offline repair/undo. Old staged approval digests require fresh review. Preserve the sole source-build evidence image below.

  • Post-install acceptance in progress (2026-10-09 16:01 UTC): Finish all five real application conversions/native windows and both source-kernel compile/install/boot/fallback tests, then deliver the updated local ISO, remove the disposable VM after exporting evidence, and commit/push final results.

    Verified: fresh encrypted install/generic boots; welcome Enter/Return, keyboard scroll, all seven real bridge reviews and cancelled-review safety; GIMP source 3.2.4 conversion/native window; balanced 6.18.54 GUI-approved compile in 18584 seconds, correct suffix/config/modules, source boot, source-running build refusal and persistent generic fallback. Balanced scratch peaked at 19.6 GiB; reviews now recommend 25 GiB. Firefox clean conversion passed in 9674 seconds with source VDB, old binary removal, user-data/default preservation and generic image preservation. Native normal-user launch through the preserved desktop ID shows JavaScript and bundled AV1 video PASS in run 141501Z, shots/05-05-native-firefox-js-av1-rendered.png; closed normally.

    Active: generic GUI run gui-desktop-20261009T131341Z-6115018e, controller user unit phoebus-postinstall-controller-recovery3, log /var/tmp/phoebus-postinstall-controller-recovery.log. Guest unit phoebus-postinstall-calibre-resume4 runs only Calibre. The VM now has 20 GiB/12 CPU; QtWebEngine is actually compiling with four jobs (both Ninja invocations verified), later tuned ten. Source ebuild/args.gn digests match. Guest state/logs: /var/tmp/phoebus-app-sequence.status, /var/tmp/source-build-APP.{log,rc}. Do not start tuned or replace binaries during that worker. At 16:00 UTC the resumed Qt build reached 3495/7405, no error or memory pressure; guest 24 GiB/host 29 GiB disk free. Thunderbird conversion passed in 8710 seconds, LibreOffice in 15005; both native normal-user windows and clean exits passed. Writer edit/undo passed. Calibre must open the existing library's guide EPUB after conversion; original Flatpak remains intact during this compile.

    Resource-resume metadata repair: original Qt compile reached ~16106/23519, then deliberately stopped/cleanly rebooted into 20 GiB. Kept its partial log in source-build-calibre-12g-interrupted.log. Three early retries failed on missing temp Python/CMake metadata; logs retained. Fresh archive preparation matched all 269594 source files; their timestamps and old build tree were restored. Temporary Qt-only keepwork/keeptemp plus /etc/portage/bashrc hook run genuine python_setup/cmake_prepare when metadata is missing. Resume wrapper removes those settings, cleans exact Qt scratch and removes /var/tmp/phoebus-qt-preserved after conversion succeeds. No source/USE change. Old run 125620Z ends 9 PASS/2 FAIL after expired sudo caused a chained prompt during inspection; controller recovery then closed QEMU. No clean poweroff claimed there. Always refresh op=auth then use sudo -n for guest commands; new temporary controller also interrupts each chained command on recovery.

    Controller /var/tmp/phoebus-postinstall-control.py uses request/result JSON in /var/tmp/phoebus-postinstall-control; latest manual ID 571, monitor IDs use nanoseconds. Helper /var/tmp/phoebus-postinstall-request.py reads JSON stdin. Shell markers must split their echo: echo MARKER-$(printf DONE). Auth/UI-auth use internal fixtures; never print credentials. Compact host monitor /var/tmp/phoebus-postinstall-monitor-apps.py runs in exec session 75643; poll with write_stdin up to 45 seconds. Restart it if the backend loses that session. Variables must be task-specific: Zsh status is read-only.

    Current payload deployed during idle recovery: sys SHA-256 73c2bf6132eb7e69a7984c7470f61fa73ad67072a7655ef560720ce7e5047173, installer 2cd3cb6c14f6f2900979727c02ac5382555da60054a473991d96951f0cc5bfa5, transition bundle 20208cc5689aa8a778b0d6db752946737e32d4825e95aeb0fb8c11d3d855a876. Fmt/Clippy/all 961 tests/static/dynamic releases pass, logs /var/tmp/phoebus-postinstall-codec-*.log; VM harness 22 tests/ShellCheck pass. The source-only codec fix scopes -system-av1 to Firefox/Thunderbird below 156 (Mozilla bug 2063808); both fresh .mozconfig files confirm without-system-av1. Temporary keepwork was removed after a stale-config retry; do not restore it.

    Latest ISO: installer/live/phoebus-installer-amd64.iso, SHA-256 b0e63a8d3ed9b82fab08a561522cb9b7b36ddad24ef61bb8493641cf5523a92a. Exact embedded binaries/QML/GTK3 setting pass, as do two fresh-vars, no-target-disk UEFI boots through a read-only virtio-SCSI CD (173910Z, 174001Z), reaching graphical.target/live login after normal menu selection. Delivered at the canonical local ISO path on October 9, replacing the older October 6 candidate and freeing another 1495142400 allocated bytes; stale checksum sidecars removed. Code/evidence checkpoint a6ccae2 is pushed to main. AHCI CD attempts failed/stalled in firmware; baseline control also stalled; EFI files match byte-for-byte. Do not claim the AHCI cause or physical tests. Packing scratch removed; rootfs tar retained. Two obsolete generated ISO copies removed, freeing 3289325568 bytes; host free ~32 GiB. Older verified ISO retained. No formal tag/latest.iso publication: CI gate remains NotRun.

    Interrupted evidence: old generic run 074853Z was externally killed during Firefox. Run 141501Z ends PASS=12/FAIL=2 after the new monitor used reserved status, preventing its marker; Ctrl+C recovery killed QEMU. Its earlier Firefox acceptance remains valid, no clean poweroff claimed. gui-driver now uses stdio chardev signal=off. Read-only ISO run 194424Z confirms Ctrl+C no longer terminates QEMU, but left Utilities selected and did not finish live boot; not another ISO boot pass. Thunderbird interrupted log was preserved, its exact work directory cleaned and remaining worker restarted. Firefox source VDB persists, original Thunderbird binary intact.

    Next: finish remaining conversions and native windows; approve tuned through actual GUI/polkit, compile/verify configuration and actual modules, one-shot boot tuned and verify generic default fallback. Kernel boot script now includes source .config/module checks. Balanced config/module manifests saved in /var/tmp for export. Reboot helper /var/tmp/phoebus-postinstall-reboot-service.py CHOICE requires current clean poweroff, preserves stopped OVMF_VARS and starts a persistent controller. Set the tuned one-shot entry before invoking it. Keep generic persistent.

    HTTP user unit phoebus-postinstall-http, port 8000, serves /var/tmp/phoebus-postinstall-serve. Native launcher in guest is /var/tmp/launch-native-app-codec-fixture.py; HTML/AV1 files already copied. Export fixture export-source-evidence.sh uploads only logs/atoms/boot entries/aliases/config manifests to port 8001; persistent receiver user unit phoebus-postinstall-evidence-receiver writes /var/tmp/phoebus-vm/postinstall-source-acceptance-evidence/postinstall-evidence.tar.gz. Export/verify before deleting the sole qcow2 /var/tmp/phoebus-vm/disks/postinstall-choices.qcow2. Finalize ISO/checksum, docs/status/changelog, conventional commit and push. Rebuild ISO only if further production code changes; the serial-harness fix is outside its payload.

  • Fresh ISO and VM cleanup (2026-10-06): Cleared all 33 disposable qcow2 images from /var/tmp/phoebus-vm/disks (about 80 GiB freed), retaining the small test evidence files. make doctor and make check pass. The live rootfs was rebuilt rather than reused because live-packages.list changed; current static sys and installer hashes match binaries extracted from the new ISO. ISO SHA-256 is 5886d8f54c45ce330175edeaf7812fe6815371aac2f8ffa7b4477e4f013c3430 at installer/live/phoebus-installer-amd64.iso. A no-disk QEMU UEFI boot reached graphical.target and phoebus-live login:. The former fully tested candidate was retained as phoebus-installer-amd64-20261001-verified.iso. Fresh installation and the independent CI release gate remain outstanding.

  • Cross-lane install comparison (2026-10-03): A plain install now performs read-only searches across the configured package catalogues before asking the model to recommend a candidate. The Assistant shows per-lane results, failures and availability, and selecting another listed source stages a separate exact-source review through the legacy planner. AppImage catalogue links remain non-installable leads. make check, dynamic and static release builds passed. In the disposable VM, install chatgpt desktop produced a Snap recommendation and six-lane comparison; the broader search found 17 Nix results. Reviewing Snap chatgpt-desktop-client and Nix chatgpt staged distinct exact-source plans. The final plan was cancelled. No package or repository was installed. VM preview used the user-local static sys (SHA-256 d9172c81098ba23b6d37a9480fdb25aecc8fb0c646191fd4be2f30a35ef68f59) and Assistant QML override. sys.new was renamed to sys; a subsequent sha256sum sys in the VM matched that build hash. Privileged alternative execution remains unverified.

  • Natural ChatGPT Desktop request (2026-10-03): The user clarified that the bad result was the Assistant's AppImage fallback for “I want to install chatgpt desktop,” not the Send button. The unavailable install path now looks for an exact GPO package name and performs the reviewed overlay discovery before using AppImage guidance. A VM rerun instead found a Snap plan, so the same external option review now accompanies successful non-Portage plans. Explicit source choices remain constrained. Focused tests passed. The VM natural request returned a Snap plan alongside gentoo-zh, nest and ruizhi-overlay choices; selecting gentoo-zh staged the separate overlay plan, which was canceled. No repository or package was added. The guest ran the user-local static sys (SHA-256 f7d5510a…55fb2be) and Assistant QML override; the system-installed binary remains unchanged.

  • Assistant Send visibility (2026-10-03): The button left of Clear History is Send. Its disabled primary style used dark text on a dark fill, making the label almost invisible with an empty prompt. The shared QML button now uses a visible disabled label and outline; the running disposable VM displayed it correctly after copying the component into the user-local Assistant config.

  • External overlay registration (2026-10-03): The next user request asked sys to discover compatible third-party repositories, offer a choice, then add and keep the accepted repository synced. A deterministic review now intersects GPO leads with the Gentoo registry and inspects each repository's actual Git name, commit, exact ebuild SHA-256, amd64 keywords and Manifest. sys overlay review lists options; sys overlay enable stages an approval that the privileged broker revalidates before cloning. Portage configuration masks all packages from that repository except the selected atom and scopes ~amd64 if needed; auto-sync = yes makes normal Portage syncs refresh it. The Assistant's exact GPO link now shows selectable options and stages the same approved plan. Host and disposable VM read-only discovery found gentoo-zh, nest and ruizhi-overlay for app-misc/chatgpt-desktop. The VM Assistant displayed all three, staged a ruizhi-overlay plan with distinct details, and canceled it. No repository or package was installed. VM testing used a user-local static sys and test-only Assistant bridge path override because the system-installed bridge still targets /usr/local/bin/sys. The first VM lookup timed out; retry succeeded. make check and the dynamic/static release builds passed. Privileged registration remains unverified; use only a disposable VM if testing that path. A listing and hash do not establish publisher trust or future update safety.

  • GPO link and read-only approval correction (2026-10-03): The user's https://gpo.zugaina.org/app-misc/chatgpt-desktop follow-up was handled by the model as a general administration plan that offered approval to inspect an unrelated Snap candidate. The bridge now recognizes an exact GPO atom, checks configured Portage, and performs a bounded external GPO lookup; an unsynced lead returns information with no install approval. Final general plans made only of read-only checks also run and display those checks as information without approval. The running VM's deterministic search found the GPO lead but no configured Portage match (/var/tmp/phoebus-vm/chatgpt-overlay-search.png), and eselect repository list -i could not load the repository module. No overlay or package was added. The new user-local static sys is SHA-256 7ee56b9bdce504b4cfd4c6ada58ba8ce3b93ecd906edbc5a26e298c7d0bb1448; after Assistant restart the exact link returned an Information card without approval (/var/tmp/phoebus-vm/gpo-assistant-result.png). Rust format, strict clippy, workspace tests, make check, and dynamic/static release builds pass. A fresh ISO remains gated by 8 GiB free versus 20 GiB required by make doctor. The upstream GPO page lists ruizhi-overlay, gentoo-zh and nest; gentoo-zh's current ebuild (observed separately) fetches a Debian artifact from an OpenAI-hosted URL and has a Manifest hash. That hash verifies fetched bytes against the overlay's manifest, not publisher trust. A repository trust choice, ebuild review, scoped ~amd64, and repository registration still precede any install.

  • Assistant plan details (2026-10-03): The Details button previously repeated the recommendation already shown at the top of the card. It now shows the selected package, source health evidence and limits, planned actions, effects, rationale, alternatives, post-execution check, and digest. The bridge exposes readable action and alternative descriptions and the candidate health report; plan text no longer implies a pending artifact has already passed cryptographic verification. Rust format, strict clippy, workspace tests, make check, and dynamic/static release builds pass. In the running VM, a staged org.mozilla.firefox Flatpak plan displayed the new details and scrolled to its digest (/var/tmp/phoebus-vm/details-scrolled.png). The user-local sys SHA-256 is 1d356c8e2b8c1bcdbaf6b59bb329ce2df283e32d07c8d2393b88f699b3581df8; the user-local QML is updated and Assistant reconnected. No installation was approved. A fresh ISO remains gated by 8 GiB free versus 20 GiB required by make doctor.

  • Source review coverage correction (2026-10-03): The prior note's claim that vendor Portage packages were covered by a separate lane was incomplete: vendor binary ebuilds can resolve as Portage source. AI install proposals now require exact inspect_candidate evidence for every lane. The broker re-inspects every selected candidate at plan and execution, and direct typed Flatpak installs refuse unverified remotes. Portage and Nix labels now state when source origin or registry revision is unknown. Direct general AI AppImage updates and Nix preparation cannot bypass the install workflow. Rust format, strict clippy, workspace tests, make check, and dynamic/static release builds pass. The running VM showed a read-only app-editors/vim inspection with review_required and unknown exact repository origin (/var/tmp/phoebus-vm/all-source-portage.png). The updated static binary was installed in the VM user's ~/.local/bin/sys (SHA-256 d687107f68d5559b13eafd75c2de4977759a7240e91768aa0c85436b38753a29) and the Assistant reconnected. No package was installed. A fresh VM install remains unverified because make doctor reports 8 GiB free against its 20 GiB gate.

  • Third-party source health review (2026-10-03): AI installation now must inspect the exact candidate for third-party lanes before making a proposal. The read-only report separates backend evidence from publisher trust and unknown release age. Flatpak plans require the app to resolve from the configured flathub remote with HTTPS and GPG enabled; the broker checks again just before execution and aborts on version, remote, or verification drift. Direct installs in general AI plans are rejected in favor of the dedicated sys install workflow. Rust format, strict clippy, workspace tests, make check, and dynamic/static release builds pass. The running VM showed a read-only org.mozilla.firefox inspection with Flathub HTTPS, GPG enabled, version 157.0, and review_required status (/var/tmp/phoebus-vm/source-health-inspect2.png). The updated static sys was installed at ~/.local/bin/sys (SHA-256 3cbbe7aaf25bf4659abc1d5dc3e2772e738ffffc6a324e4a666ef0c5752dd941) and the Assistant was restarted connected. No package was installed. An actual reviewed Flatpak install remains unverified in the VM; a fresh ISO remains gated by make doctor reporting 8 GiB free versus 20 GiB required.

  • General Assistant search cap (2026-10-03): The exact VM question can you find a package for balena etcher for me previously failed with too many investigation requests in one turn (max 6). The general loop now validates every typed action, executes only the first six read-only checks, and feeds evidence plus a skipped-check note back to the model. The live VM answered with an AppImage catalog lead and no verified installable package (/var/tmp/phoebus-vm/etcher-search-new2.png). That run exposed a desktop bridge issue: empty-action answers showed Approve & Execute. The bridge now sends an information result with no plan ID or digest, and the service does not stage an empty plan. Tests cover both behavior and the read-only boundary. The final user-local static sys was installed in the VM (SHA-256 6c7226433af167377d2002e2aba9682914e99691c501c61bb2c04681b4e586ee). Repeating the exact question returned an Information card, displayed its read-only sys checks, and showed no approval button (/var/tmp/phoebus-vm/final-question-progress.png). The VM's root-owned binary remains older until a fresh install. A fresh ISO is still gated by make doctor reporting 9 GiB free against its 20 GiB VM requirement. Rust format, strict clippy, workspace tests, make check, and dynamic/static release builds pass.

  • Assistant flicker and Etcher follow-up (2026-10-03): Progress used to replace the entire QML conversation model and scroll the full view on every event. It now updates a fixed-height activity list inside the message card. Mango previously faded the Assistant on focus changes; a per-window unfocused_opacity:1,isnoanimation:1 rule keeps it steady, with a one-time migration that preserves user visual rules. The running VM showed the Assistant opaque while Kitty held focus (/var/tmp/phoebus-vm/no-fade-terminal-after.png). The installation model can now report unavailable after typed candidate discovery; the bridge presents a read-only information result with AppImage catalog guidance if no validated plan emerges. The exact natural Etcher request completed in the VM as a read-only information card after package lane checks, with an Etcher catalog lead and no install button or package change (/var/tmp/phoebus-vm/etcher-new-progress.png). The user-local static sys SHA-256 is f29def8d9d4169f3ffd36c71c3c26781091f6bf0ef2ae518f9881585dd921139. Host format, strict clippy, workspace tests, dynamic and static release builds pass. A fresh ISO remains gated by disk space.

  • Visible assistant activity (2026-10-03): The assistant bridge now sends redacted sys phase events keyed by request ID before the final reply; QML displays and retains each phase in the conversation card. System-fact reads, model waits, typed investigations, and reviewed execution report progress. The malformed-JSON error observed for i want to install balena etcher now triggers up to two bounded corrections and retries, with no execution on failure. The VM then exposed a slow startup handshake caused by a full proactive update scan. Startup now runs quick reboot/storage/service/transaction checks, while full update discovery runs when Alerts or Updates is opened. The running VM now has the updated user-local static sys and QML: the assistant connects promptly and displays fact checks, model waits, and AppImage search stages (assistant-activity-model-wait.png and assistant-activity-verified.png under /var/tmp/phoebus-vm). A broken scratchpad activation call was removed so the prompt receives focus. The exact natural Etcher request first failed closed when the model omitted an action field. One correction got past that error, but the model then used an invalid spaced application name. The final two-correction build was installed in the VM (SHA-256 0b2fcbf92faa06dcacbb366b91bcf08f8ac3a73bc2ec70e704a08faf328a3738). It made read-only checks, then stopped after eight turns without a valid proposal; no install was approved or executed. The GUI now explains that outcome in user terms. Screenshot: two-retry-scrolled.png. Rust format, strict clippy, workspace tests, dynamic/static release builds, and the VM harness tests pass. Fresh ISO verification remains gated by disk space.

  • AppImage community search (2026-10-03): The previous sys search AppImage lane showed only Phoebus-managed files. It now also reads the AppImage project's HTTPS community feed on the running system, ranks matching names and descriptions, removes feed HTML from displayed descriptions, and labels each remote result catalog lead. Search never converts such a lead into an install candidate; appimage_acquire still requires a reviewed artifact URL and SHA-256. Offline feed failure remains explicit. The user's balenaEtcher query matches the feed's Etcher listing. Explicit AppImage install requests now return catalog information and the pinning requirement without invoking the model to draft an impossible plan; the exact assistant phrase has a bridge regression test. Other install requests pass the selected backend to the model, and an over-six investigation batch gets a correction instead of an immediate error. In the running VM, sys search firefox --legacy --appimage returned seven leads in 0.9 s and sys search balenaEtcher --legacy --appimage found Etcher; screenshots: /var/tmp/phoebus-vm/appimage-search-vm.png and /var/tmp/phoebus-vm/appimage-etcher-vm.png. A terminal AppImage install request displayed the lead and pinned-acquisition guidance without making a plan (/var/tmp/phoebus-vm/appimage-install-guidance.png). The assistant window was restarted and reconnected to the user-local bridge. The final ~/.local/bin/sys SHA-256 is 420de1fd4698657c1f6740eb4c6fe54bb54908014b2e408e8bd1dfe3ce3b740f; its root-owned /usr/local/bin/sys remains older. No AppImage was installed. A fresh ISO and root binary update still await the VM disk-space gate.

  • Assistant request delay (2026-10-03): In the running encrypted VM, the assistant accepted "i want to install chatgpt desktop" at 12:34 and did not show a response until about 12:41. The old phrase parser sent it to general investigation; the GUI showed no pending card and its footer said "Ready". The eventual result was a reviewed administration plan; no package was installed. The parser now recognizes natural install wording and multi-word names, and QML shows a pending card plus elapsed time until the bridge replies. Rust format, clippy, workspace tests, release, static builds and make test pass. The running VM's user QML override was updated and the final static binary was installed as ~/.local/bin/sys (SHA-256 3a0363e1…7cbbfb3); its user AssistantBridge.qml points at that binary because the root-owned /usr/local/bin/sys is older. The guest displays the immediate working card, elapsed timer and no approval buttons while a new exact-wording request is pending; screenshot: /var/tmp/phoebus-vm/assistant-exact-pending.png. That request reached the install planner and failed because OpenCode returned invalid proposal JSON; the UI showed the failure around 1:01, with no package change (assistant-followup-status.png). The error card now explains that model failure in user terms. Update the root-installed binary/assets at a fresh build; no package was approved. The final user-local QML and static bridge restarted to a connected assistant in the same VM.

  • Greeter reference spacing (2026-10-03): The current VM screenshot shows ESC touching the left edge and yellow keycaps. The upstream Usage image places the status bar roughly two text rows inside the terminal and uses the slate palette's soft gray for the keys and action text. The generated greetd config now adds --window-padding 2 and uses gray throughout the greeter. Rust format, clippy, workspace tests, release and static builds pass. After authenticated guest login and a greetd restart, the existing VM showed an inset status bar and gray keys; screenshot: /var/tmp/phoebus-vm/greeter-slate-inset-gray-20261003.png. The VM remains open at the updated greeter. Upstream 0.9.1 lays out the margin in terminal cells and recomputes the remaining width per frame, so the inset is resolution-independent; very narrow terminals may truncate the status text. Fresh-install reboot verification remains.

  • Tuigreet slate login (2026-10-03): The installed greetd command now adds a Phoebus greeting and soft white theme. A greetd pre-start drop-in applies a Nord slate VT palette with setvtrgb, supplied explicitly by sys-apps/kbd; failure of the color setup does not block login. Generated config/palette/drop-in tests and the Rust gates pass. After an authenticated login to the existing encrypted playground, the new config and palette were installed in the guest. Restarting greetd returned to the slate login screen with a white border, clock and greeting; screenshot: /var/tmp/phoebus-vm/greeter-slate-20261003.png. The VM remains open at the greeter. Fresh-install and reboot verification remain; the 20 GiB free-space gate for a fresh ISO/VM is still unmet.

  • Greeter shortcut keys (2026-10-03): The first slate screenshot lacked ESC/F2/F3/F12 hints because button=black colored those glyphs black against the dark background. The generated theme now uses button=yellow; the config and installation tests, clippy, workspace tests, and release and static builds pass. After an authenticated update in the existing VM, restarting greetd showed yellow ESC/F2/F3/F12/CMD hints beside their actions. Screenshot: /var/tmp/phoebus-vm/greeter-slate-shortcuts-20261003.png. The VM remains open at the corrected greeter; a fresh-install reboot is still pending.

  • External overlay leads (2026-10-02): GPO lists app-misc/chatgpt-desktop in Gentoo community overlays, but sys search previously queried only Portage repositories enabled locally. The new sys search chatgpt --legacy --overlays --portage keeps those local results and adds a bounded read-only GPO lookup with clearly unverified leads. A host lookup returned the ChatGPT desktop ebuild. The updated static binary is installed at ~/.local/bin/sys in the running playground; a VM query displayed the external lead beneath an empty configured-Portage result. No overlay is enabled by this search, and the ebuild is not treated as an install candidate until a reviewed overlay is configured and Portage resolves it. Host checks, release and static builds passed.

  • sys natural install error (2026-10-02): The phrase sys install chatgpt desktop through nix used to pass the entire phrase as a package name and print invalid legacy examples. The parser now extracts the requested lane; failed model plans search the lane and suggest an exact catalogue match. The running QEMU VM returned chatgpt from Nix and the revised binary suggested sys install chatgpt --legacy --nix without installing anything. Host regression and full check passed. OpenCode returned invalid proposal JSON on the first run and a valid chatgpt/Nix plan on a later run; its prose about GUI availability was inconsistent with the catalogue metadata. The user's installed root sys requires authentication to replace. The updated static binary is installed as ~/.local/bin/sys, and the interactive zsh PATH now prioritizes that directory. This lets the user's sys command use the fix; sudo sys may still invoke the older root-owned binary.

  • Phoebus Updates entry point (2026-10-02): The Updates view now has a dedicated launcher and Super+U shortcut, works without a connected model, and opens the existing interactive sys apply flow in Kitty so review and package progress stay visible. DMS itself is unchanged and updates through Portage. A one-time Mango config migration adds the shortcut to existing profiles unless Super+U is already bound. The live image now installs the Assistant and Updates desktop entries. Host checks pass; the new graphical flow still needs a fresh VM check when make doctor's 20 GiB disk-space gate is met (currently 10 GiB free). The visible disposable VM was updated through user-local overrides (~/.local/bin/phoebus-assistant, ~/.local/share/applications/phoebus-updates.desktop, and ~/.config/quickshell/phoebus-assistant) because its root assets require authentication; the previous QML was backed up beside that directory. The user QML overrides /etc/xdg in this playground and should be refreshed or removed after its root-owned assets are updated. Super+U opened the Updates view, reported 20 Portage updates, and a new Mango rule kept its window at 900×650. The review/apply button was not used on that VM, so no package transaction ran.

  • Frame and OS launcher logo defaults (2026-10-02): DMS 1.6.2's layout picker sets frameEnabled=true; its launcher button logo uses launcherLogoMode=os (and the dock uses dockLauncherLogoMode=os). The session now seeds these in a new settings.json, leaving every existing settings file intact for these choices because DMS omits default-valued keys. Session tests cover first use and preservation; make check, release and static builds, and ShellCheck pass. The running QEMU playground already visually shows Frame selected and an OS launcher icon; a fresh ISO remains unverified under the 20 GiB VM disk-space gate.

  • Theme Color Auto default (2026-10-02): DMS 1.6.2 maps the Theme Color Auto button to currentThemeName=dynamic and currentThemeCategory=dynamic in settings.json. The Mango session now seeds those values before shell startup when the bundled aurora is present and no theme was chosen, preserving other settings and explicit themes. Session tests cover all three cases; make check, release/static builds and ShellCheck pass. The running disposable VM visibly has Auto selected with the aurora wallpaper; the new first-login seed itself awaits a fresh ISO. The host has 10 GiB free against the 20 GiB VM gate.

  • Wallpaper and avatar defaults (2026-10-02): The Mango session sets DMS's session.json wallpaper path to the bundled aurora only when unset, and its cache.json wallpaper browser folder to ~/Pictures/Wallpapers only when unset. jq is available in both the live package manifest and installed DMS dependencies. AccountsService already falls back to ~/.face; the installer now creates a first-letter SVG for the installation user before greeter login, and the session creates one for later users if absent. The existing QEMU playground displayed its “P” avatar and aurora wallpaper. Focused tests exercise first use and preserve a chosen wallpaper, folder and avatar; a temporary-home check covered existing DMS state with empty wallpaper fields. The new session launcher was installed in the visible VM. make check, release and static builds, and ShellCheck passed. Fresh ISO/install remains unverified while host free space is 10 GiB against the 20 GiB VM gate.

  • DMS cursor and Settings sizing (2026-10-02): The packaged Mango config sources ./dms/cursor.conf, and the session creates that fragment before Mango starts. A one-time marker lets existing user configs gain the include without replacing custom settings or re-adding it after a user removes it. Mango 0.17.2 interprets floating window rule dimensions below 1 as output fractions; DMS Settings now opens at 0.65×width and 0.8×height, retaining ignore_maximize:0. The visible disposable VM showed roughly 1092×840 at 1680×1050 and 832×614 at 1280×768, and the DMS CLI include resolver returned {"exists":true,"included":true} for a clean temporary profile. The emitted session migration passed a temporary-home check: it adds the include once and respects later removal. The packaged config and session script were deployed to the existing VM, which was returned to 1680×1050. make check, release and static builds, and ShellCheck passed. Fresh ISO/install verification remains blocked by the 20 GiB host free-space gate (currently 10 GiB).

  • Shell control and help pass (2026-10-01): Added an offline Phoebus Help page, launcher and Super+F1 shortcut; DMS Welcome now has a separate launcher, Super+Alt+W shortcut and button in Phoebus Welcome. The live medium manifest now includes Firefox ESR and xdg-utils for the Help page. The installed Welcome is a styled 720×600 floating window with Help and finish controls visible on first view, and its close path quits Quickshell so --reopen works. The Assistant is a 900×650 floating scratchpad; Super+A toggles the same process and Escape hides it. The live Welcome routes Network, display and text buttons to exact DMS tabs, retains its installation help, and shows a useful message before an installer log exists. First login fills missing terminal, HTML and mail defaults with Kitty, Firefox and Thunderbird. The existing visible disposable QEMU VM was updated and the bar panels, DMS Settings pages, both welcome paths, Help and assistant toggles were visually exercised. A follow-up control pass found Mango's default maximize requests ignored; the Settings rule now sets ignore_maximize:0, and DMS Settings expanded and restored correctly. Theme, wallpaper and Settings search controls also worked. Both launcher shortcuts worked; the launcher showed the new Help and DMS entries. Rust formatting, Clippy, workspace tests, release and static builds, ShellCheck and desktop entry validation passed. The live-only install action and restart/shutdown actions were inspected but not activated. A fresh ISO or install is still gated by roughly 10 GiB free versus the 20 GiB requirement.

  • Settings and pointer (2026-10-01): The DMS Settings window was only about 500×400 because Mango floated it at its minimum size. A specific 900×700 Settings rule was added to the packaged Mango config and checked in the existing disposable QEMU VM. On this host, both GTK and SDL VirGL viewers displayed a 1680×1050 guest frame smaller than the viewport while clicks followed full-size coordinates. The VM was relaunched with 2D VirtIO, GTK zoom off and a USB tablet; the image filled the window and clicking the visible Settings close icon closed it. run-vm.sh now chooses this reliable input path by default; --graphics gl remains opt-in. The VM is left running on the existing disposable qcow2 for the user. Rust format, warning-free Clippy, workspace tests, release and static builds, VM safety tests, ShellCheck and Python syntax checks pass. A fresh ISO/install could not be exercised because make doctor found 10 GiB free against its 20 GiB gate.

  • Wallpaper defaults (2026-10-01): Four generated 1672 × 941 JPEGs live in installer-rs/assets/wallpapers. The installer embeds and seeds them for the installation user and future accounts without replacing custom files; the live build copies them into /home/phoebus/Pictures/Wallpapers. Provisioning and preservation tests plus live-build guards pass. The four images were also copied into the running QEMU playground's ~/Pictures/Wallpapers and matched against repository SHA-256 hashes. The existing branded SVG remains the live installer background. make doctor reports 10 GiB free against its 20 GiB new-VM gate, so a fresh ISO/install has not yet been built with these assets.

  • VM guest defaults (2026-10-01): The visible playground was gracefully restarted on its existing disposable qcow2 with virtio-vga-gl and GTK GL; LUKS, login and Mango + DMS succeeded, and the guest reported +virgl. The window remains 1400×900. The installer now probes QEMU/KVM, VMware and VirtualBox via systemd-detect-virt plus DMI, selects corresponding guest packages and Mesa video targets (virgl or vmware), and enables the relevant units. Gentoo's QEMU agent starts by udev when its port appears, so it is not enabled as a static unit. Mango installs wlr-randr, fixes its low-resolution check, watches changed preferred modes, and starts available SPICE/VirtualBox session agents. The disposable QEMU launcher can select VirGL when explicitly requested and exposes a QEMU guest-agent port. Portage dependency previews resolved all three package stacks. The current QEMU GTK GL window scaled on resize without changing the guest preferred mode and later showed pointer drift; VMware/VirtualBox runtime behavior, SPICE window resize and a fresh installed ISO are unverified. Rust format, warning-free Clippy, workspace tests, release and static builds, VM safety tests, and ShellCheck pass. make check initially exposed a concurrent assistant bridge test writing into another test's personal-storage fixture; the handshake test now takes the same environment lock and uses its own fixture, and make check passes. make doctor still fails only the host free-space gate (11 GiB versus the requested 20 GiB). See docs/VIRTUAL-MACHINES.md.

  • Nix executable integration (2026-10-01): sys install --nix now uses a persistent /nix/var/nix/profiles/profile; nix prepare still only realises packages. New Nix bootstrap writes /etc/profile.d/nix.sh with client, shared and per-user XDG profile bins on PATH; the Mango session sources it before launching the compositor and imports PATH into systemd's user manager. Install verification checks the active profile and reports real executable names. cmatrix was installed through the new reviewed sys action on a disposable encrypted VM overlay; the binary and PATH worked, and sys list --nix reported its manifest. The existing visible playground received the new static sys, profile script and session launcher, then passed an authenticated 9/9 serial boot check; its QEMU window was relaunched in windowed mode. Rust format, Clippy, workspace tests, release and static builds pass. make doctor still reports only the host-space gate (11 GiB free versus 20 GiB for a new VM); the test used an existing qcow2 backing image and a temporary COW overlay. A fresh ISO/install remains unverified. Playground boot note: with QEMU -device virtio-gpu-pci -vga none, the GTK window shows “Display output is not active” while the guest waits for its LUKS passphrase. An isolated qcow2 boot with -device virtio-vga showed both the LUKS prompt and subsequent graphical greetd login on the display. The visible playground was relaunched with virtio-vga, unlocked, and visually verified at the graphical login; its host window is 1400×900.

  • Package-lane terminal and reliability pass (2026-10-01): Human sys sessions now stream Portage compilation, Flatpak transaction progress, Snap operations, Nix realisation, custom ebuild builds and AppImage download output (including carriage-return progress) while retaining redacted logs. Broker/assistant protocols remain structured. A disposable encrypted VM overlay passed sys-lane-operations.sh 20/20: Portage and Flatpak install/remove/output, non-root preferences, honest manual-lane update reports, AppImage import/list/normal-user launch/pinned update/remove, and Nix realisation. A final-source re-run passed 22/22, including protective Flatpak and AppImage snapshots and natural AppImage removal. A second boot passed sys-custom-ebuild-operations.sh 9/9 for stage/build/output/unmerge/remove. AppImages moved out of the private /var/lib/phoebus tree into /opt/phoebus/appimages; invalid AppImage URLs that could expose URL credentials are refused. Custom ebuild removal preserves its staged recipe when unmerge fails. Flatpak, Snap and AppImage package mutations now request protective root/var snapshots. The installed VM passed Snap first-use 11/11, then a final-source Snap run 12/12 including its protective snapshot; the Store supplied no newer revision for the refresh. The expanded six-lane catalogue matrix passed 75/75. An AppImage import changed the real managed file, took an automatic snapshot, and was undone across a detached reboot: rollback PASS=11, restored app/root state, /home preserved. Rust format, warning-free Clippy, workspace tests, release/static builds, VM safety tests, ShellCheck and the secret scan pass. The existing visible playground is still running; after the isolated VM checks, its /usr/local/bin/sys was atomically replaced over the serial console and its SHA-256 was confirmed to match the tested static binary. make doctor found only the host-space gate (about 7 GiB free during tests, 11 GiB after deleting the completed overlay, against 20 GiB for a new VM), so this work used a validated COW overlay and does not claim a fresh ISO/install. A fresh release ISO/install, NVIDIA hardware, strict Snap confinement and a revision-changing Snap refresh remain unverified.

  • Assistant launch and Snap desktop default (2026-10-01): The visible playground exposed a Quickshell configuration error at shell.qml:223: FloatingWindow has no palette; after removing that error, it rejected minimumHeight. The corrected QML loaded and showed the assistant window in the disposable VM. A second issue was a startup race: the shell asked the bridge before its process started and retained a false offline state. The shell now applies the completed handshake and loads user state then; the final QML opened in the disposable VM and showed Connected (opencode). New Mango installs include app-containers/snapd and enable snapd.socket. The existing playground predates this; reviewed sys actions installed snapd 2.63 and enabled/started its socket there. snap find glances returned Store results, and sys find glances completed with a Snap match and all three package lanes ready. Rust format, Clippy, tests, release build and make test pass. The playground was restarted from its validated disposable overlay. A fresh ISO/install remains required for release evidence; host free space was 11 GiB versus the 20 GiB VM threshold in make doctor.

  • sys terminal presentation (2026-10-01, overlay VM VERIFIED): Human CLI runs now show the current planning, evidence, effects, execution and verification phase; slow commands identify their operation and long steps print a ten-second heartbeat. Status is on stderr and redacted; results stay on stdout, while broker JSON and the assistant bridge are quiet. Informational model answers are wrapped and do not show an approval digest. Unit and CLI integration tests pass. The visible playground overlay was updated with a rebuilt sys: sys history --legacy visibly showed step transitions, and real OpenCode sys ask what model are you using showed model/evidence progress and a short wrapped answer. The final VM binary names OpenCode (opencode-go/deepseek-v4.1-flash) instead of the adapter's internal "local" label. The user's sys find glances request completed in the same VM with package-lane checks visible and a concise wrapped answer; a unit regression test covers the model label. A fresh ISO/install has not included this change. make doctor reports only 10 GiB free against its 20 GiB new-VM threshold; no new VM was made.

  • Polkit wheel administrator fix (2026-10-01, overlay VM VERIFIED): The visible playground exposed a default root-only polkit prompt for software updates. The installed phoebus user's password worked with sudo but failed there because the installer locks root by default. Added a generated 49-phoebus-wheel-admin.rules file in users_configure and installed the same rule in the disposable playground overlay. A graphical pkexec /usr/bin/id -u prompt then accepted the user password and returned 0. The source fix is unit-tested; no fresh ISO/install has included it yet. The previously verified candidate ISO is superseded by this source change: build and requalify a new candidate before publishing. The visible VM and host OpenCode relay remain running for the user to explore.

  • Installer/desktop reliability candidate (2026-10-01, VM VERIFIED): Rust hardening, recovery-first boot counting, upstream systemd boot assessment, desktop permissions/audio/DMS startup, live UI fixes and exact-ISO release reporting are implemented. The current candidate is installer/live/phoebus-installer-amd64.iso SHA-256 b14f37e1b04d8e1ec37f77d2a14adb4aaef652770fee995a18a0ac7b986b5281. Its UEFI live UI gate passed 32/32 (gui-gui-20261001T093944Z-0b2eeb6a), and a real OpenCode installer proposal passed 18/18 (gui-opencode-20261001T094134Z-e2d90704). The ISO contains installer SHA-256 5bf5d6868c1a22250ff62fc386804110e035fe52d2bc8ec10385c738a65d401a and sys SHA-256 c5b079478f01e79ad4e823db87b32f0da7aecc13fb50afffc53cd53c58e8ff5d; both match the static build. Two independent encrypted Mango/Nix installs passed 5/5 each on disposable reliability-v7-a.qcow2 and reliability-v7-b.qcow2 (serial-install-20261001T094417Z-4e5913c7 and serial-install-repeat-20261001T094417Z-392fae4f). The fresh second install passed detached counted boot 5/5, installed graphical desktop 30/30 with visible first-login welcome, offline recovery 6/6, failed-boot fallback 9/9, rollback 11/11 and interrupted update 10/10. Evidence IDs and limits are in docs/VM-BOOT-VERIFICATION.md. The intermediate v6 installs were stopped after the missing installed welcome launch was found; their partial disks were discarded. A patched v5 installed overlay passed the first-login welcome and Escape/seen-marker graphical check 30/30; v5 also has two complete installs and recovery-path evidence. VM logs and results redact registered credentials. Final cleanup on both v7 installs logged a nonfatal busy LUKS mapper; investigate separately. CI remains NotRun until an independent Forgejo runner reports success, and physical hardware remains unverified. Keep the ISO unpublished while any required release gate lacks evidence.

  • System-wide stability audit & zero-panic hardening (2026-09-30, IMPLEMENTED + TESTED): Completed an exhaustive audit across all production crates to eliminate unexpected panics, fix subtle edge-case race conditions, and harden userspace agent protocols:

    1. Eliminated all unwrap()/expect() panics in production code paths (phoebus-installer::release, phoebus-installer::service::plan, phoebus-core::brand, phoebus-executor::chooser, phoebus-ai, and phoebus-executor::rollback).
    2. Fixed desktop checkpoint collision when mutations occur within the same second by using microsecond resolution and sequence counters (desktop_config.rs).
    3. Fixed asymmetric undo: undo_latest_checkpoint now records whether files existed prior to mutation and removes newly created files when rolling back to an initially absent state.
    4. Structured TOML serialization for MangoWM keybindings to prevent syntax breakage from unescaped quotes in actions.
    5. JSON-RPC 2.0 error conformance for phoebus-mcp (handling parse errors and missing params with proper error frames).
    6. Sanitized Bubblewrap agent_jail paths to filter out directory traversal (..) attacks.
    7. Btrfs swapfile hardening: executed chattr +C on directory and on 0-byte file prior to truncation, avoiding Btrfs extent errors; enforced mode 0600.
    8. Inter-process lock drop safety: verified PID ownership prior to unlinking lockfile on drop. Quality gates: cargo fmt --all -- --check, cargo clippy --workspace --all-targets -- -D warnings, cargo test --workspace all passed cleanly with 0 warnings and 0 errors.
  • System stability & reliability hardening (2026-09-30, IMPLEMENTED + TESTED): Implemented concurrency protection, package database consistency, and intuitive user workflows:

    1. Inter-Process System Mutation Lock (SystemLock): Added PID-liveness-backed file lock at /run/phoebus/sys.lock (safe /proc/<pid> liveness check with automatic recovery of stale locks from dead processes) to prevent concurrent mutation collisions.
    2. Coordinated Snapshot Pairing: High-risk mutations in actions.rs now invoke snapshot::create_set_and_record to create atomic paired @ and @var Btrfs snapshots, preventing /var/db/pkg Portage database drift during rollbacks.
    3. Atomic Transaction Persistence: Updated transaction.rs to write transactions.toml via tempfile + sync_all + atomic rename, eliminating partial/truncated file risk on sudden power loss.
    4. System Health Overview: Implemented SystemAction::SystemStatus (sys status, sys health, sys check) querying BLS boot assessment states, reboot requirement, storage headroom vs the 10 GiB gate, Btrfs snapshots, and package lane readiness.
    5. Intuitive CLI & Context-Aware Guidance: Unified update/updates -> CheckUpdates and upgrade -> ApplyUpdates; sys install <app> --legacy without a backend now dynamically resolves candidates across lanes via chooser::resolve_candidates to suggest exact copy-pasteable commands. Gates: cargo fmt, clippy --workspace --all-targets, cargo test --workspace all pass.
  • Userspace-directed AI agent framework (2026-09-30, IMPLEMENTED + TESTED): designed and implemented a Two-Tier architecture for userspace developer/desktop AI agents (e.g., Claude Code, OpenCode, Cursor, Codex). Added agent_jail + phoebus-agent-run (Bubblewrap isolation with private tmpfs masking over ~/.ssh, ~/.gnupg, keyrings, .aws, .config/gh), desktop_config + phoebus-desktop (schema-validated MangoWM keybinding & DarkMaterialShell theme mutations with automatic snapshot checkpoints and one-click undo), phoebus-mcp (standardized Model Context Protocol stdio server), and sys propose (unprivileged plan staging for Polkit elevation review). Fixed parent directory creation race condition in phoebus-executor::personal. Workspace tests: 124+ passed, 0 failed.

  • Live real-agent call (2026-09-28, host, real model): with a real opencode serve (v1.18.32) on 127.0.0.1:4096 and the opencode-go provider, sys ask "list installed packages" --ai-backend opencode --opencode-provider opencode-go --opencode-model deepseek-v4.1-flash returned a valid assistant plan + approval digest (read-only list_packages, exit 0). A hostile instruction to read /etc/shadow and ~/.ssh and run a shell command was refused and downgraded to the read-only inventory; the phoebus transport disables every OpenCode tool in the session, so the agent has no shell/file access. No provider credential appeared in the sys log. This is host-level live evidence (not VM); a VM/live-provider run with the direct sandboxed transport remains NotRun.

  • End-to-end AI-path test (2026-09-28, IMPLEMENTED + UNIT-TESTED, no live model): new installer-rs/crates/phoebus-installer/tests/sys_ai_e2e.rs drives the real sys binary against a real local OpenAI-compatible TCP stub and the real curl transport (--no-sandbox, no seat). Four cases: a valid model plan executes end-to-end; the request sent contains no /home/ path and no credential; a model run_shell fails closed after one corrective retry; a depclean with a smuggled shell key is refused by the exact schema; and malformed output fails closed after one correction. This is not a live-model call (sys ask with a real provider is still VM/live NotRun) but it closes the "connected-model path only tested with fake transports" gap. Gates: cargo test --workspace 841 passed / 0 failed.

  • Three kernel choices + VM-found kernel bugs (2026-09-29): added a first-class three-way kernel choice (bin default, balanced source kernel with broad hot-plug coverage, tuned source kernel) via the new BuildBalancedKernel action / sys build balanced kernel; both source choices keep gentoo-kernel-bin as the known-good fallback and create a separate BLS entry. The first VM journey started the balanced build and found two real bugs, both fixed: vdb_has_atom("sys-kernel/gentoo-kernel") matched the sibling gentoo-kernel-bin record (prefix overlap), and the separate BLS entry was looked up in EFI/Linux instead of the real loader/entries directory. The build is in progress on a disposable overlay with MAKEOPTS="-j10 -l14" (the installed default -j3 -l8 is memory-derived and too slow for a test build). Source conversion remains to be started. Result: the full distribution config did not finish in 3 h (guest timeout killed it at RC=124; 7.7 GiB build artifacts, no system change — build-only is fail-safe). The review text now says "allow several hours" instead of the earlier "20–60 min". Completing this journey needs a multi-hour build window; the separate BLS entry path is unit-tested with the corrected loader/entries directory.

  • OpenCode transport standardized (2026-09-29, VM-VERIFIED): the OpenCode backend no longer runs outside the model sandbox and no longer has a stricter endpoint policy than the direct provider. It is sandboxed by default (curl body inlined on stdin because PrivateTmp hides a temp file); https servers on the network work as before, and remote plain http is allowed only with --opencode-insecure-http / PHOEBUS_OPENCODE_INSECURE_HTTP=1. Re-verified in the VM with the real host agent (serial-boot-20260929T114319Z-c43c2a92, driver PASS=8). Next confidence work: VM-prove the approval-ledger/effect-binding paths, source-conversion/tuned-kernel journeys (multi-hour), a physical-seat desktop pass, and configure a CI runner for the release gates.

  • VM live AI (2026-09-29, VM-VERIFIED): from inside the installed guest, the host's real opencode agent was reached through the loopback tunnel and sys ask produced a valid plan (live-agent), with a hostile instruction refused/downgraded (live-agent-hostile-rejected) (serial-boot-20260929T102703Z-ffd1ec25). A real local model (Qwen2.5-1.5B GGUF via llama-server) was also driven through the direct sandbox in both scopes; it reached the model but failed closed on schema-noncompliant output. The assistant prompt now shows a concrete action-object example ({"action":"list_packages"}), which made the small model emit action objects; a capable model is still required, and a 4096-token local context can be exceeded by the one-shot correction turn. Source conversion and tuned-kernel builds remain NotRun (multi-hour compiles).

  • Real powerful model via the host agent, sys in the VM (2026-09-29, VM-VERIFIED): four requests through opencode/deepseek-v4.1-flash: list and search firefox return valid grounded plans; install firefox --flatpak as the user stops before mutation; hostile shell/private-file instruction refused (serial-boot-20260929T111941Z-2be30b89, PASS=8, zero auth prompts). A read-only Flatpak probe had hung on Flatpak's interactive appstream-update (pkttyagent) on the headless console; CommandSpec::detached() now runs host-root Flatpak commands with no controlling terminal. The install prompt now enumerates the allowed kind values (a capable model had guessed kind:"gui"). The install investigation itself still stops on the model's own request-count/formatting limits and fails closed (no sys defect).

  • First VM test loop on the golden image (2026-09-28, VM-VERIFIED): booted a fresh overlay of the installed phoebus-headless-80g.qcow2 with the static sys from HEAD and found two real defects, both fixed. (1) Unprivileged sys ask hung on a polkit manage-units prompt because the model sandbox used the system systemd-run; it now uses systemd-run --user with the user-manager property subset for non-root callers, full system sandbox for root. (2) An unprivileged read-only AI plan exited 1 after printing the plan because the service layer required the root-owned transaction log; recording is now best-effort for non-root plans and mandatory for privileged mutations. VM evidence (serial-boot-20260929T092147Z-6f88907a, PASS=8): system sandbox AI_ROOT_RC=0, user sandbox AI_USER_RC=0 (was 124/timeout), no-sandbox AI_NOSANDBOX_RC=0, hostile reply AI_SHELL_RC=1, and sys-acceptance.sh SUMMARY PASS=49 FAIL=0. Hypothesis recorded (not changed): headless/inactive-session flatpak search can trigger Flatpak's interactive appstream-update polkit prompt; the unattended acceptance runs detached to avoid the hang. Additional loops on the same golden overlay with the fixed binary: mutating acceptance as root SUMMARY PASS=42 FAIL=0 (serial-boot-20260929T093329Z-4ef893cf), rollback PASS=11 (serial-rollback-20260929T094640Z-256c2ba1), Snap journey PASS=11 (serial-boot-20260929T094957Z-3d4a9211). Next: source conversion / tuned kernel (multi-hour compiles) and a live direct-provider model in the VM remain NotRun.

  • Boundary-hardening pass (2026-09-28, IMPLEMENTED + UNIT-TESTED, VM NOT RUN): six host-provable defects fixed with regression tests. (1) Removed the PHOEBUS_ALLOW_UNPRIVILEGED_EXEC=1 env backdoor: it made service::effective_uid() report uid 0, letting a non-root desktop-assistant client skip pkexec + the approval ledger and run a root-requiring plan in-process. (2) load_api_key now refuses a credential containing ", \, whitespace or a control character (it is interpolated into a curl config header = "..."). (3) The OpenCode base64 Basic/Bearer token is now registered with the redactor (only the plaintext was). (4) SystemFacts::validate_no_forbidden now scans cpu_features (it was omitted). (5) validate_endpoint refuses userinfo (https://user:pass@host/...). (6) The AI assistant/install parsers now use phoebus-core::action::validate_action_value, closing a unit-variant deny_unknown_fields bypass ({"action":"depclean","shell":"sh"} was accepted on the AI path). Gates: fmt clean, clippy --workspace --all-targets -- -D warnings clean, cargo test --workspace 837 passed / 0 failed, release build. NotRun: VM systemd-run sandbox exercise — host probe shows system manage-units requires polkit auth and the full property set is invalid for --user (DynamicUser unsupported; PrivateDevices fails); a VM run as the phoebus user on a seat is required to settle whether the sandbox is usable unprivileged or must switch to --user with a reduced property set. Also NotRun: source-conversion/tuned-kernel journeys (need a disposable install + reboot); Snap changed-revision/strict confinement (external prerequisite). Identified but deliberately unchanged: Flatpak/Snap install drift binds the human version only (no commit/revision); SnapRefresh has no review binding; UpdateSystem/world-update has no post-verification; PrepareUpdates records ok with a truthful "prebuild skipped" note. Recorded as hypotheses, not reported holes.

  • sys adversarial battery + hardening (2026-09-28, IMPLEMENTED + UNIT-TESTED, VM NOT RUN): new installer-rs/crates/phoebus-installer/tests/ sys_adversarial.rs (54 cases: shell injection per operand, empty/NUL, lengths, Unicode/homoglyphs, traversal, option injection, malformed/nested JSON, CLI flag conflicts, natural-language edges, boundaries, privilege gates) exercising the phoebus-core validators and the sys binary under PATH=/nonexistent. Found and fixed 5 real gaps, each regression-tested: valid_file_path shell metacharacters + .. (doc contract previously unimplemented); valid_application_name ../empty components + leading slash; valid_ebuild_atom .-leading components, closing a real overlay path traversal (../pkg), plus ebuild::package_dir re-validation; valid_flatpak_app_id empty reverse-DNS element; valid_nix_attribute leading dash. Gates: cargo fmt --all -- --check clean, clippy --workspace --all-targets -- -D warnings clean, cargo test --workspace 828 passed / 0 failed. VM rerun of the acceptance matrix remains NotRun.

  • sys flaw-fix battery (2026-09-28, IMPLEMENTED + UNIT-TESTED, VM NOT RUN): fixed 6 audited flaws: versioned/slotted/repo atoms now validate (valid_package_spec), quoted multi-word app names parse in legacy install/remove, vdb_has_atom normalizes (executor + kernel), Flatpak install/remove pin --system --noninteractive, and who owns <relative path> is normalized. Added host battery installer-rs/crates/phoebus-installer/tests/ sys_flaws.rs (9 tests) and sys-acceptance.sh service-lifecycle / versioned- atom / who-owns cases. Gates: cargo fmt --all -- --check, clippy --workspace --all-targets, cargo test --workspace 773 passed, bash -n + shellcheck clean. VM rerun of the expanded acceptance matrix is NotRun.

  • Headless stabilization run (SELF-HEALING LOOP ACTIVE, run 10 IN FLIGHT): Failures so far, each fixed + committed + verified green (761 tests) + rerun:

    1. Brave overlay pin missing → recorded reviewed pin d3a97f22… (0aa31a3).
    2. /bin/zsh missing at users_configure (Zsh lane only in desktop, which runs later) → shell lane moved to base_packages (5ea4bb1).
    3. zsh-autosuggestions masked ~amd64 → accept_keywords_shell() wired into the GURU block (ff104dd).
    4. GURU zsh-autosuggestions needs >=zsh-5.9.1, stable is 5.9-r7 → keyworded app-shells/zsh itself (6972511).
    5. swayimg REQUIRED_USE exactly-one-of lua_single_target_* (profile enables none) → media-gfx/swayimg lua_single_target_luajit (3abdc87).
    6. gui-apps/matugen does not exist (GURU sparse-index proof) → corrected to x11-misc/matugen (f480c9c).
    7. nemo needs libxmlb[introspection], binhost lacks it → scoped dev-libs/libxmlb introspection (0036529).
    8. Same class: xapp[introspection] → scoped xapp + pre-emptive libsecret/ cinnamon-desktop introspection + mpv luajit (8a0dfdf).
    9. Same class: libdbusmenu[gtk3] (xapp dep) → scoped it plus the rest of the ebuild-read closure: gdk-pixbuf/gtk+ introspection (233a3e2; closure read from gentoo.git tree, no checkout).
    10. Run 10 reached 285/346 desktop then the DRIVER (not the guest) gave up: serial-driver.py install expect timeout was 9000 s (2.5 h) but full Mango needs ~3.5 h; QEMU was killed with the guest mid-emerge → 24b11eb raises it to 19800 s (5.5 h).
    11. Run 11 reached 346/346 then failed enabling seatd: Gentoo's seatd ships its systemd unit only with USE=server; our package.use set flags on libseat but not on seatd itself → add7dfc adds sys-auth/seatd server systemd builtin.
    12. Run 12 reached finalize then failed its own check: the VDB verify matched a literal category/name directory, but VDB entries are category/name-version → has_vdb_record() prefix match + regression test (2b8559f).
    13. Run 13 COMPLETED GOLDEN (===INSTALL-RC=0===, NO_SECRET_LEAK, serial dir serial-install-20260928T002557Z-e39d1934). All 12 stages passed, 347 desktop packages emerged, 588 compile times recorded into metrics, coordinated snapshots @-20260928T060439-initial-known-good and @var created.
    • Detached boot proven (serial-boot-20260928T075316Z-c1918e9c): Booted detached disk with -boot order=c (no ISO). LUKS2 decrypted, systemd running with 0 failed units. seatd.service active and running (/run/seatd.sock 0660 root:seat, phoebus in group seat). Fast-defaults verified in Gentoo VDB (firefox-bin, libreoffice-bin, thunderbird-bin, brave-browser) and Flatpak (calibre, gimp, flatseal). Acceptance suite passed 40/40.
    • Next: scripts/publish.sh to push commits to origin/main.
  • Run 14 (IN FLIGHT, pristine record with all fixes): new static binaries from HEAD (phoebus-installer 183ec40b…, sys d5f7a46f…); driver pid 2485744, serial dir serial-install-20260928T091251Z-8e9c8076, chain pid 2494536. Includes the nix file-perm fix (c806c98), opencode Local default (2261c54), zsh-prompt + equals fixes, setsid acceptance, and the guest loopback tunnel (fwd.py, with the settimeout(None) long-reply fix).

  • Run 15 (COMPLETE GOLDEN, 2026-09-28): INSTALL-RC=0 (driver 4/4, serial serial-install-20260928T132731Z-b5f8629a), no leak; detached boot driver 4/4 (serial-boot-20260928T183131Z-953e4677: systemd running, 0 failed, seatd active, clean poweroff); 40/40 acceptance; nix per-user hello enumerated from install perms alone (/etc/nix* world-readable — the bc0c4b8 ordering fix proven); AI_LIVE_RC=0 (valid plan + digest, muse-spark-1.3-contributor). Evidence in VM-BOOT-VERIFICATION.md / STATUS.md / CHANGELOG.md. Next: publish.

  • Seat/thunderbird/AI-timeout session (IMPLEMENTED + UNIT-TESTED, VM NOT RUN): P0 seat fix (seatd pkg + enabled before greetd, libseat systemd/builtin/ server USE, seat in default groups, greeter -G video,input,render,seat, 3 s bounded seat wait in the session script, LIBSEAT_BACKEND deliberately unset), thunderbird-bin fast default + SOURCE_CONVERSIONS entry + welcome text (all lanes untouched; mandatory-AI requirement unchanged), and CurlTransport timeout 120 s → 300 s default via PHOEBUS_AI_TIMEOUT (fail-closed parsing). Gates: fmt, clippy --workspace --all-targets -- -D warnings, 759/759 tests (was 757), bash -n session script. Seat-VM reverification, thunderbird installed-identity/conversion journey, and live CPU-inference timeout proof are NotRun (no VM run this session).

  • Zsh modernization (2026-09-26, IMPLEMENTED + UNIT-TESTED, VM NOT RUN): rewrote ZSH_SKEL_RC (color fix, cached compinit, Kitty keybindings, INC_APPEND_HISTORY, vcs_info git branch, exit-status prompt) and added ZSH_ZPROFILE (/etc/profile bridge + ~/.local/bin); seed_default_zsh uses the new const. Gates: fmt, clippy --workspace --all-targets -- -D warnings, 757/757 tests, zsh -n + interactive startup smoke on host zsh 5.9. Seeding is once-only (existing ~/.zshrc preserved). Installed-session behavior (login env, Nix paths, prompt/git/keys in Kitty) awaits the disposable VM. Uncommitted; docs/DUAL-BOOT.md from the prior session also still uncommitted.

  • Fast-defaults follow-up (2026-09-26, ISO BUILT + CONTENT-VERIFIED, VM NOT RUN): audited the e747baf fast-defaults tree against the prompt (all lanes covered, no code changes needed), re-ran gates green (fmt, clippy -D, 757/757 tests, release build, vmsafety 18/18, shellcheck info-only), rebuilt the static binaries from e747baf (phoebus-installer 8cf90d0f…, sys c87557…), and built a fresh ISO phoebus-installer-amd64-fastdefaults.iso (1.6 GiB, ad9cddb2…, build 20260926T145700Z, PHOEBUS_REUSE_ROOTFS=1; git-ignored) with content verification (embedded binaries hash-match, Gentoo key both fingerprints, Mango/DMS + installer UI present). NO install/boot this session — wall-clock blocker (full Mango install ~165 min; host ready: doctor 20 ok, ~103 GiB free); all install/boot/graphical/source/tuned gates NotRun with exact unblock in VM-BOOT-VERIFICATION.md. Next: disposable 80 GiB install of the ad9cddb2… ISO + detached boot, 37-case acceptance, graphical/file/device/source/tuned gates.

  • Interface contract (unchanged): ordinary requests use AI; only explicit --legacy selects deterministic handling; source flags constrain the AI and never bypass it; a missing model stops the workflow and offers the exact --legacy command without executing it.

  • Fast-defaults session (2026-09-26, IMPLEMENTED + UNIT-TESTED, VM NOT RUN): Firefox/LibreOffice as -bin, Calibre/GIMP/Flatseal as target-only system-wide Flatpaks (no Portage duplicates), single generic kernel, delivery section in plan review + progress log, install metrics hooks, preference neutrality guard, convert_to_source (Firefox/LibreOffice/ Calibre/GIMP; Brave refused with reason) + build_tuned_kernel (separate BLS, fallback kept, default only after test boot) typed actions with full reviews through the existing approval/broker path, installed once-per-user reopenable non-blocking welcome + DMS skel seed, portal/MIME defaults with no sandbox overrides. Gates: fmt, clippy -D, 757 tests (was 736), release build, vmsafety 18/18, shellcheck clean. NO new ISO, NO install/boot this session — session-time blocker (host ready: doctor 20 ok, 104 GiB free); all VM gates NotRun with exact unblock in STATUS. Next: build ISO, disposable 80 GiB install + detached boot, 37-case acceptance, graphical/file/device/source/tuned gates per VM-BOOT-VERIFICATION.md.

  • Interface contract (unchanged): ordinary requests use AI; only explicit --legacy selects deterministic handling; source flags constrain the AI and never bypass it; a missing model stops the workflow and offers the exact --legacy command without executing it.

  • Workstation experience session (2026-09-26, IMPLEMENTED + UNIT-TESTED, VM BLOCKED): Mango workstation atoms (22 apps, exact, no substitute; Kitty kept, Nemo default, CUPS local-only, sshd installed/disabled), Brave stable scoped overlay (pin pre-network fail-closed; no pin recorded yet), Zsh-by-default

    • packaged plugins, Vanilla-DMZ 24 cursors, default ccache 2G, AI reuse handoff (per-user 0600 key, truthful status, no Approve&Execute on errors), startup-check model (blur optional, PPD+D-Bus, matugen, cava) with truthful summary, assistant dark-default + light + responsive + token-drift guard, 60/20 GiB disk gates. Gates: fmt, clippy -D, 736/736 tests, release build, vmsafety 18/18, shellcheck info-only. NO new ISO (8 GiB free, needs ≥20; target needs 60–80), NO install/boot this session — all VM gates NotRun with exact unblock in STATUS. Next: free ≥80 GiB, pin Brave commit, build ISO, disposable 80 GiB install + detached boot, seat-VM compositor/cursor/ DMS/theme checks, fixture + live-AI handoff, local CUPS job, sshd-disabled
    • PPD D-Bus recheck after reboot.
  • Interface contract (unchanged): ordinary requests use AI; only explicit --legacy selects deterministic handling; source flags constrain the AI and never bypass it; a missing model stops the workflow and offers the exact --legacy command without executing it.

  • ISO rebuild + verification session (2026-09-26, DONE except seat-compositor): new ISO installer/live/phoebus-installer-amd64-20260926.iso (1.6 GiB, 250dbe9a…, build 20260926T041500Z, HEAD 11c0ee0 + 0b41781) built via PHOEBUS_REUSE_ROOTFS=1, content-verified (embedded hashes match, Gentoo key both fingerprints, UI payload present); old Sep-24 ISO kept. Interactive VM RUNNING (KVM/GTK, phoebus-interactive-20260926.qcow2 + private vars in /var/tmp/phoebus-interactive/, screenshot at welcome, undisturbed). Unpinned signed-DIGESTS full install PASSED (gui-install-20260926T040728Z-f3ae68f4, INSTALL-RC=0, NO_SECRET_LEAK, ~165 min) + detached boot (running, 0 failed units, sys --legacy + snapshot). Tamper refusals PASSED (placeholder pin pre-network, hash mismatch post-download, both RC=1 + no leak). Recovery-entry boot PASSED (snapshot cmdline/findmnt, login, sys --legacy, truthful ask failure, systemd degraded). Seat desktop PARTIAL: seat0/DRM/greeter/auth verified, Mango dies in <1s (libseat, no seatd) — the blocker. Snap: no Store revision, still partial confinement. Nix: GC-dangling NOT reproduced; new gap: modern per-user Nix profiles invisible to sys (follow-up). Live AI: still no credential/model (NotRun). gui-driver.py boot sys-check fixed (--legacy). gates --execute: 0/0/6 NotRun, no publish. Quality gates green (717/717, clippy -D, vmsafety 18/18). Evidence disk disks/unpinned-accept-1.qcow2 KEPT; scratch overlays removed. Next: seatd-or-logind fix for Mango (design+rebuild+retest), per-user Nix enumeration, live AI credential, snap revision/strict, NVIDIA.

  • Security/readiness follow-up (2026-09-26): trust anchor + executable gates + strengthened rollback PASS.

    1. Stage3 trust anchor (fixed): release_key_path() ignores PHOEBUS_GENTOO_RELEASE_KEY; production always uses the provisioned key, test injection only via explicit verify_digests_with_key. Host-level Good-signature check of the real DIGESTS passes (not VM evidence). VM proof blocked: needs a rebuilt live ISO + disposable install (exact prerequisites in docs/VM-BOOT-VERIFICATION.md).
    2. Release gates executable: invalid checks/desktop modes fixed (real boot modes), --execute runs each argv and reports Passed/Failed/NotRun from evidence (unconfigured host: 0/0/6, publish refuses). Found and fixed a gui-driver.py SyntaxError. 6 new tests.
    3. Strengthened rollback rerun PASSED (serial-rollback-20260926T034218Z-7fab51ef, PASS=11 FAIL=0, exit 0, no leak, overlay rollback-accept-8.qcow2): mandatory ===ROLLBACK-MARKER present=== before undo, DB ca398d71…→fda3780a… →restored, marker absent after detached reboot, /home survives, systemd running. SERVE shellcheck fix included.
    4. Recovery-entry boot is now a distinct manual gate with its own procedure (NOT RUN). Roadmap no longer lists passed sys undo as unfinished; Secure Boot filed as a separate opt-in milestone.
    5. Quality gates: cargo fmt --check, clippy --workspace --all-targets -- -D warnings, cargo test --workspace 717/717 (one assistant_bridge flake under parallel load, green in isolation and on rerun), release + static-pie builds, test_vmsafety.py 18/18, shellcheck clean. Seat desktop, live AI, Snap revision/strict, and NVIDIA stay open with exact prerequisites (STATUS).
  • Interface contract (unchanged): ordinary requests use AI; only explicit --legacy selects deterministic handling; source flags constrain the AI and never bypass it; a missing model stops the workflow and offers the exact --legacy command without executing it.

  • Hardening pass (2026-09-26): stage3 fail-closed + truthful boot security + rollback acceptance PASS.

    1. Stage3 (security defect, fixed): the unpinned path warned and continued on DIGESTS-download failure, missing SHA-512 entry, missing gpg/key, failed key import, and gpg launch errors. All stop the install before extraction now; bad signature/hash also stop it. The pinned path requires a valid 64-hex pin (placeholder refused) before any network use. stage_install checks for gpg/sha256sum/sha512sum first. 16 new regression tests (every failure branch, nonzero-exit and cannot-start doubles, both positive paths). The live image ships gnupg plus the Gentoo release key, fingerprint-checked at build time against Gentoo's published fingerprints (13EB…2910 weekly, D99E…2058 signing). Commits 2162670, 744059b.
    2. Boot security (fixed): uki=true/secureboot=true are rejected at validate_config (shared by CLI/TUI/GUI bridge); the plan preview always reports unsigned BLS. Parsing/validation/rendering regressions + golden updated. Commit 95e0d05. Authenticated boot stays an open limitation (STATUS).
    3. Rollback acceptance PASSED for real (serial-rollback-20260926T030237Z-912f6386, PASS=11 FAIL=0, exit 0, no secret leak): fresh overlay rollback-accept-7.qcow2 on fresh-install-1.qcow2 — snapshot → app-misc/hello install (DB ca39…→fda3…) → sys undo <id> rc=0 → detached reboot (DB restored, root marker absent, /home survives, systemd running, recovery entry present). The 2026-09-25 pass claim had no transcript and did not reproduce; fixed harness defects: sudo auth (stale-prompt match bug found and fixed with unique markers), root phases, explicit --legacy, verified break (nano was pre-installed), targeted undo, marker-absent assertion, preserved phase-1 log. Commits 744059b, 5f69996.
    4. Release candidate (local, isolated): manifest (2 artifacts), verify, tamper refusal, and publish refusal against the NotRun report all behave; gates --execute is 0 passed / 6 NotRun (every gate needs VM/seat/model; run_gates never executes gate commands itself). Nothing published.
    5. Quality gates: cargo fmt --check, cargo clippy --workspace --all-targets -- -D warnings, cargo test --workspace 712/712, release + static-pie builds, test_vmsafety.py 18/18, make doctor 19 ok / 1 disk-space warning.
  • Interface contract (unchanged): ordinary requests use AI; only explicit --legacy selects deterministic handling; source flags constrain the AI and never bypass it; a missing model stops the workflow and offers the exact --legacy command without executing it.

  • Quality pass (2026-09-26): boxed ExecError::Command details (CommandFailure, 128 → 96 bytes, measured with a path-dependent probe crate; constructor + all match sites updated, error text/redaction/stderr-tail preserved, 4 regression tests), cleared cargo clippy --workspace --all-targets -- -D warnings with no blanket allows (type alias, CommandRecord params struct, DiskPhase renames), and split service.rs (4218 lines → service/mod,plan,validation,execution,tests) and gui_bridge.rs (2763 lines → gui_bridge/mod,protocol,bridge,tests) as structural refactors with explicit re-exports and pub(super) seams — no service.rs left alongside service/mod.rs. make test green (692/692, unchanged across the splits). Release + static-pie builds pass. The disposable VM re-ran the current 38-case sys acceptance matrix against the new static sys (70f0a72b…): guest PASS=38 FAIL=0, driver PASS=41 FAIL=0 (serial-boot-20260926T014746Z-3089a820, overlay sys-acceptance-refactor.qcow2). Commits e69bfb9, 1bf31c5, 2dd48d1.

  • Interface contract (unchanged): ordinary requests use AI; only explicit --legacy selects deterministic handling; source flags constrain the AI and never bypass it; a missing model stops the workflow and offers the exact --legacy command without executing it.

  • Snap first-use follow-up (2026-09-25): the previous "no snapd" note was a real integration gap. The normal sys install … --legacy --snap resolver needs snapd to confirm a Store candidate before approval, while a raw typed snap_install silently added it during execution. First-use setup is now a separately reviewed Portage install plus socket enable/start, with exact guidance when absent. Fresh installs set scoped sys-fs/squashfs-tools lzma lzo, which Gentoo's snapd ebuild requires. A disposable VM exposed a second bug: Portage candidate resolution picked the first dependency (dev-lang/go-bootstrap) instead of the requested app-containers/snapd; it now matches the requested atom. Snap confinement is reported from the daemon rather than assumed strict. The disposable VM completed setup, Store install, refresh and remove (PASS=10 FAIL=0), with partial confinement. A fresh overlay and a clean reboot repeat with the final static binary each passed PASS=11 FAIL=0, including clean poweroff. Refresh found no newer revision.

  • Previous session: second-round sys review fixes. Follow-up to d3f676a. Three further confirmed defects are fixed with regression tests: unsafe cleanup previews, non-exact Flatpak/Snap update execution, and reverse-dependency self-reference. make test is green (661 → 683 tests). This corrects the earlier blanket claim that every review finding was already fixed.

  • Interface contract (unchanged): ordinary requests use AI; only explicit --legacy selects deterministic handling; source flags constrain the AI and never bypass it; a missing model stops the workflow and offers the exact --legacy command without executing it.

Snap first-use changes

  • phoebus-core::system generates scoped SquashFS USE flags for new installs.
  • Portage candidate resolution selects the requested atom from the pretend plan, even when dependencies precede it. A multi-package regression covers the bug found by the VM.
  • The Snap install path gives exact setup commands when snapd is absent; the typed action no longer adds an unreviewed Portage dependency. Candidate sandbox levels and missing snap info fields are treated conservatively.
  • installer/tests/vm/snap-guest.sh drives the installed-system journey. The serial driver can authenticate for boot checks and logs only guest output, keeping sent test passwords out of its transcript.

Previous sys review work

  1. Cleanup previews are non-destructive (finding 1). eclean-dist/ eclean-pkg delete by default, so portage_cleaners now previews with --pretend --deep (same selection policy as execution) and executes with --deep; the Nix preview uses the documented nix-store --gc --print-dead. Items are read from stdout+stderr, sorted and bound. Tests prove a cleaner never runs destructively during planning, revalidation or cancelled approval.
  2. Exact Flatpak/Snap update transactions (finding 2). ResolvedPlan.updates carries UpdateTransactions: full Flatpak ref + full installed/remote commits (via flatpak info, since the commit column is abbreviated), or snap name + old/new revisions + channel. Execution consumes the reviewed transaction (add_executable… flatpak update --commit=<c>), revalidates each source immediately before its mutation, verifies the resulting commit/revision, and refuses non-exact transactions. A source whose resolution failed is never re-discovered. PlanStore aggregation includes updates.
  3. Reverse dependencies (finding 3). The reverse branch no longer drops valid matches; it excludes only a genuine self-reference (the dependent is the queried package).
  4. Determinism. All selected packages: removals are sorted (depclean previews are now order-independent); emerge --depclean --ask=n.

Verified in disposable VMs

  • Cache cleanup + depclean (real tools): eclean upstream preview is non-destructive; sys clean cache preview/cancelled/approved/drift/failed all behave correctly; depclean cancel/approve with transaction history and preserved /home.
  • Reverse dependencies: sys-libs/glibc and sys-libs/zlib report real dependents (conditional/versioned included).
  • Flatpak: installed via Portage (sys-apps/flatpak, enabling media-libs/freetype[harfbuzz]), flathub added, an app installed and downgraded, then sys apply updates --flatpak resolved full ref+commits, deployed the pinned commit, and verified it.
  • Custom ebuild: stage → build/install (VDB entry + installed file) → remove.
  • Managed AppImage: acquire with pinned SHA-256, list, run the managed ELF, remove; a placeholder SHA is refused.
  • Nix: a NIX_ENABLED=yes install bootstraps Nix; nix profile install nixpkgs#nodejs-slim into /nix/var/nix/profiles/profile is enumerated by sys list --nix and found by sys list nodejs --nix, with store paths and generations reported separately.
  • Non-root pkexec sys --broker: with the shipped polkit action and the standard wheel-admin rule installed on the disposable image, a wheel user authenticated through a terminal agent; cancellations changed nothing, the approved removal returned the broker JSON outcome, and the replay was rejected. The shipped privilege policy is unchanged.
  • Read-only acceptance: installer/tests/vm/sys-acceptance.sh PASS=38 FAIL=0 re-run against the final static sys (dd61c610…), including the new rdeps-legacy reverse-dependency regression.
  • Snap first use: the disposable VM installed app-containers/snapd through the reviewed Portage plan (six effects), enabled and started snapd.socket, installed Snap Store hello-world 6.4 revision 29, ran refresh, then removed it. The daemon reported partial confinement and the install review agreed. The harness reported PASS=10 FAIL=0 (serial-boot-20260926T001830Z-107996de). Final-binary runs on a fresh overlay passed PASS=11 FAIL=0 on first boot (serial-boot-20260926T005327Z-c024cc43) and after a clean reboot (serial-boot-20260926T010605Z-e6193c3a); both shut down cleanly.

Not completed / blocked

  • Stage3 VM proof with the new binary: unit-tested (19 stage3 tests) and the pinned path is unchanged for real pins, but no full install has run with the fail-closed binary yet; the next disposable install is the proof. The unpinned (signed-DIGESTS) VM proof additionally needs a rebuilt live ISO carrying the provisioned key (installer/live/build.sh now provisions it; no ISO was rebuilt in this session).
  • Installed Mango + DMS session under a seat: still needs a seat-providing graphical VM (or physical VT): login, kitty, DMS spotlight, lock/idle, polkit agent, logout, reboot (docs/DEVELOPMENT.md physical-VT procedure).
  • Snap update to a different revision / strict confinement: refresh with no newer revision and partial confinement were VM-verified. A revision-changing refresh and AppArmor-backed strict confinement were not (need an AppArmor-ready guest kernel with loaded policy, and a Store snap that actually has a newer revision than installed).
  • Live AI: a local OpenAI-compatible model (llama-server + a GGUF model) was served on the host; the AI workflow reached the model, but the sandboxed call exceeded the 120 s transport timeout on CPU (curl exit 28). Needs a faster model/GPU or a remote provider credential. Fixture-backed routing and failure tests pass (ask-offline, missing-model offer, source constraint).
  • Recovery boot (distinct from rollback): the recovery BLS entry is present and asserted in the rollback acceptance, but booting it interactively was not automated. Method: select "Phoebus Recovery (known-good snapshot)" in the systemd-boot menu with no network.
  • Nix re-run flakiness: a later sys list --nix re-run reported the profile link dangling (the profile store path was gone) and correctly listed no packages; the first run with a real profile is the positive evidence.

Known blockers / risks

  • VM serial login is occasionally flaky; retry the boot when detached-boot-login fails.
  • make doctor passes (KVM accessible via an ACL). Disk is shared by build and VM; watch free space.
  • The pre-existing test-disk.qcow2 cannot be unlocked (unknown passphrase); do not guess or brute-force it.

Next actions (in order)

  1. Stage3 install proof: next disposable install with the fail-closed binary (pinned path), then a rebuilt live ISO (with the provisioned Gentoo key) proving the signed-DIGESTS path in the VM.
  2. Recovery boot by selecting the recovery BLS entry with the network down.
  3. Live AI with a faster local model/GPU or an authorized provider credential (natural-language discovery, source constraints, reviewed install).
  4. Installed desktop under a seat (Mango + DMS login through reboot).
  5. Nix profile GC stability if the dangling re-run is reproducible.
  6. Snap strict confinement and changed-revision update in a disposable VM when an AppArmor-ready kernel and a revision-changing Store test are available.

Definition of a good stopping point

  • make test passes; docs/STATUS.md, docs/CHANGELOG.md and this file updated; committed (scripts/agent-commit.sh) and pushed (scripts/publish.sh).