Make a passing CI/release result mean the checks actually ran. - .forgejo/workflows/ci.yml is fail-closed: it provisions the pinned toolchain or fails, then runs one 'make ci' (fmt-check, clippy, workspace tests, VM-harness safety tests, release build, static installer build). - Pin stable Rust 1.93.1 in installer-rs/rust-toolchain.toml; run cargo gates with --locked for reproducible dependencies. - phoebus-release gates now records passed/failed/blocked/skipped/not_run (only passed is a pass), executes real host gate commands, and imports the existing VM-harness results.json instead of a competing harness. Gates with no automation are always blocked. - Bind gate reports and harness evidence to a candidate name+SHA-256 (--candidate; PHOEBUS_VM_ARTIFACT[_SHA256]); stale or unbound evidence is failed, never passed. - publish refuses unless the report is candidate-bound and all-passed, the manifest verifies the artifact and a detached signature verifies; atomic publish preserves and restores the previous release on failure. - vmsafety.Results records the verified artifact digest; 3 new Python tests. - 20 new Rust regression tests; ADR 0012 and docs updated. Verified locally: make ci (364 Rust tests, 19 harness tests, release and static builds) exit 0; static binary is static-pie linked. CI runner, VM and model gates remain unexecuted here.
7 KiB
7 KiB
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.
- Last updated: this session (2026-09-22), CI and release verification
hardening (branch
codex/harden-ci-release-verification). - Current focus: Made CI and release verification mean what they claim.
(1)
.forgejo/workflows/ci.ymlis fail-closed: it provisions the pinned toolchain or fails, and runs onemake ci(fmt-check, clippy, workspace tests, VM-harness safety tests, release build, static installer build).installer-rs/rust-toolchain.tomlpins stable 1.93.1; all cargo gates use--locked. (2)phoebus-releasegate statuses are nowpassed/failed/blocked/skipped/not_run(onlypassedis a pass); host gates execute real commands; VM/seat/model gates import the existing VM harnessresults.json(--harness-results). (3) Gate reports and harness evidence are bound to a candidate name+SHA-256 (--candidate;PHOEBUS_VM_ARTIFACT[_SHA256]), so stale evidence cannot authorize another build. (4)publishrefuses unless the report is candidate-bound and all-passed, the manifest verifies the artifact and a detached signature verifies;atomic_publishrestores the previous release on failure. - Last verified (this session):
- 364 Rust tests + 19 harness tests green;
cargo fmt --all -- --checkandcargo clippy --workspace --all-targetsclean (exit 0);make ciexit 0. - Release build and static installer build succeed; the static binary is
static-pie linked(4.99 MB). - New regression tests (20 Rust + 3 Python) cover: absent prerequisites are
blocked; a failed host gate isfailedwith diagnostics; explicit skips and--onlyare never passes; manual gates areblocked; stale or unbound harness evidence isfailed;publishrefuses on not-run gates, unbound/stale reports, missing manifest, manifest mismatch, missing or failing signature, and a failed publication leaves the previous release intact. - CLI smoke:
gates(allnot_run),gates --execute --only rust-fmt(executes and reportspassed), harness-evidence import (installpassed), andpublishrefusal on an unbound report.
- 364 Rust tests + 19 harness tests green;
- Not yet verified this session: no Forgejo Actions runner is registered,
so the workflow is unexecuted. The VM, graphical-seat and live-model gates
are
blockedhere (no disposable VM/model in this environment).no-model-recovery,updates-and-rollbackanddesktop-sessionhave no automated harness yet and are alwaysblocked, so a full publication is not possible until they are automated (intentional: no false pass). - Public repo: https://git.phoebus.club/phoebus/phoebus_os
(remote
origin, SSH port 2222).
Next actions (in order)
- Complete the graphical install with a real model: download a small local
OpenAI-compatible model (or supply a 0600 provider key), then run
gui-driver.py guithrough disk/account/review/ERASE/install/rebooot twice on fresh qcow2 disks and record the passing journeys. The current automated journey passes to the AI gate only. - Installed Mango desktop: install the default profile in a graphical VM and verify the panel/launcher/terminal/audio/network/lock/idle, the dwl fallback, and the welcome/installer absence on the installed system.
- Rollback VM acceptance: deliberately break an update, run
sys undo, reboot detached, verify/var/db/pkgmatches/, then boot the recovery entry with no network or model (ADR 0007). - Narrow the live broker privilege: replace the live-only
NOPASSWD: ALLrule for the GUI bridge with the smallest practical command-scoped rule once a safe sudoers form is verified (ADR 0011 context). - Physical-seat desktop: run
phoebus-sessionon a real VT/seat and checkphoebus-menu,phoebus-lock,phoebus-idle, kitty, XWayland, the polkit agent, logout/reap and reboot (docs/DEVELOPMENT.md). - Register a Forgejo Actions runner and confirm
make ciruns green there; no runner is registered, so.forgejo/workflows/ci.ymlis unexecuted.make cipasses locally, but that is not CI verification. - Automate the three blocked release gates:
no-model-recovery,updates-and-rollbackanddesktop-sessionhave no harness scenario yet, so they are alwaysblockedand a full publication is impossible. Add harness modes (or reviewed manual evidence) and re-runmake release-gates.
Known blockers / risks
- No Forgejo Actions runner is registered, so
.forgejo/workflows/ci.ymlis unexecuted;make ciis green locally only. Registering anact_runnerwith a rustup toolchain is the external prerequisite (ADR 0012). - A full release cannot be authorized yet:
no-model-recovery,updates-and-rollbackanddesktop-sessionhave no automated harness, so the gate report is never all-passed. This is deliberate until they are automated. - No live DeepSeek/OpenAI key and no local model are available; the full GUI install past the AI step and two fresh-disk graphical installs are NOT verified (normal installation requires a connected model). The AI step is verified as a blocked prerequisite and with a fixture transport.
- Installed Mango (GURU + compiled Quickshell/DMS) is not VM-verified; nor is the dwl fallback session.
- Rollback and the AI sandbox are implemented and unit-tested but not VM-verified. NVIDIA is not verified.
- A separate physical computer for the OpenCode transport test is not available (BLOCKED); a forwarded-hop + unauthenticated-LAN-refusal host-local test was done instead.
- The GUI bridge reaches Quickshell via
Process+/usr/bin/sudo -n; it relies on the live user's passwordless sudo. The live sudoers rule is stillNOPASSWD: ALL(live-only); narrowing the broker permission is future work. - Harness artifacts live under
/var/tmp/phoebus-vm/(outside git):results.json,screenshot-manifest.json,report.html, screenshots.
Environment notes
- Dev host: Ubuntu 26.04. QEMU/OVMF/xorriso/rustfmt/clippy/ImageMagick present.
- Debian-Live base ISO is cached at
/home/dominic/debian-live-standard.iso(matchesinstaller/live/base.sha256); the appliance ISO builds without root. - Local QML iteration:
/tmp/opencode/run-qml.sh <config-dir>runs a config on a nested Mango over the host X11 using the built rootfs libraries, andgrimcaptures it. The sanctioned verification is the disposable VM. - VM harnesses:
installer/tests/vm/gui-driver.py(graphical live ISO:gui/install/boot),appliance-driver.py,serial-driver.py.DISKmust be a disposable qcow2; never point it at/dev/*. - Serve the install fixtures from
/home/dominic/vmlab/servewithpython3 -m http.server 8000 --bind 127.0.0.1(reachable as10.0.2.2:8000).
Definition of a good stopping point
make testpasses;docs/STATUS.mdanddocs/CHANGELOG.mdupdated;docs/HANDOFF.mdupdated; committed and pushed (scripts/agent-commit.sh+scripts/publish.sh).