Platform Support Watch-List (living document)

Saved host CA trust and stale clients (BI-F89C7FBD)

The supported Node gate launch (pnpm run pregate, including its native gate entry) reads existing authorized endpoint/trust through the installer client-env resolver. A desktop client can remain open after setup persists the Windows User environment or macOS/Linux managed environment. Its process can therefore lack NODE_EXTRA_CA_CERTS. Launch resolves the saved setting and starts a fresh Node process when the selected roots were not loaded at startup. Child gates inherit that environment. Shared native MCP and source OAuth requests also apply the selected bundle per request alongside Node’s effective default roots.

Trust precedence is explicit DPF_MCP_CA_FILE, explicit DPF_PKI_TRUST_BUNDLE, process NODE_EXTRA_CA_CERTS, install bundle, saved user bundle, then the existing managed root when present. A selected missing/invalid/expired bundle stops with its own diagnostic; it never silently falls back to a different CA. Without a managed bundle, ordinary public CA verification remains in effect. No certificate is harvested from the endpoint or installed by launch.

A bounded verified TLS handshake precedes host guards, authentication and lease admission. It sends no HTTP request or credential. Hostname mismatch, expired server certificates, unverified issuer and connectivity errors identify separate remedies. HTTP401/403 remain authentication/authorization failures, and a confirmed drain retains the existing quiescence wait. Transport/prerequisite failure exits 88 with no verdict about the diff. Repair the named configuration or managed PKI problem before re-running the gate; do not disable verification. A slow quiescence read remains inconclusive after transport verification; the existing authoritative late admission still governs writes.

Windows uses the User environment through PowerShell 5.1+. macOS/Linux use the existing managed environment reader through Bash, without sourcing saved text. Windows launch and fixture tests establish local behavior; adapter simulations and shell twin tests do not establish native macOS/Linux execution.

Contributor preview mount identity on Docker Desktop (BI-0ACB97B5)

The Windows host records the admitted worktree as a drive-absolute path while Docker Desktop reports its bind source as /run/desktop/mnt/host/<drive>/.... The preview supervisor and Linux-hosted sign-in issuer share the same path comparison. Translation applies only to a Windows admission; another drive, another worktree, relative paths and traversal fail closed. Source SHA, lease, loopback port, environment class and current ownership checks remain required. Regression fixtures cover both verification adapters; a failed initialization or binding check never earns a served UX pass.

The full local-CI runner also reuses the verified native shell adapter for its child environment. Windows fixtures that invoke sh must not inherit WSL or depend on an agent manually extending PATH. A missing compatible shell is an unavailable prerequisite (exit 88), not a failing product test. The parent environment, worker limits, test deadlines and lease fences stay unchanged.

Literal Windows package-manager arguments (BI-263055E3)

The shared host command resolver launches pnpm/Corepack JavaScript entrypoints with Node and a literal argument array; native pnpm executables run directly. It honors the selected installation and PATH precedence, supports npm global and workspace bin layouts, and refuses unsupported shims without a shell fallback. The observed process runner uses the same boundary with shell mode disabled. Local-CI version pinning supplies a JavaScript forwarding entrypoint so percent expansion, quotes and command separators remain argument data. POSIX executable selection is unchanged.

Capability reconciliation on release installs (BI-5ACBAC50)

The portal’s Linux container is not the host platform. Explicit v2 runtime reconciliation filters service eligibility using the signed installation’s Windows, macOS or Linux identity on both server and packaged actuator. Preserve the recorded release Compose overlays and canonical environment file through initial execution and signed recovery. Base-only Compose can reintroduce source builds on a source-free consumer install. The actuator’s packaged catalog remains bound to the signed catalog hash. Source fixtures cover the platform matrix and release context; live host application requires a qualified containing release.

Host process inventory before admission (BI-6DE5A3B6)

Windows process inventory uses explicit UTF-8, terminating PowerShell errors, and a bounded producer. Both Windows and POSIX responses must be complete and have valid process identities. Producer failure, timeout, output exhaustion and malformed responses stop before resource admission as host_inventory_unavailable (exit 88, command unrun). Diagnostics contain reason and byte count, never raw command lines or parser excerpts. Retry the original governed command after collection recovers. The intermittent live parsing cause remains unproven; these safeguards address independently reproduced failure handling.

Host claim transport and Windows command launch (BI-091A2DCF)

Host admission uses the existing lease-call allowance rather than the generic 10-second probe budget. Whole-response deadlines include trickling replies. Timeout recovery reconciles the exact immutable claim key before one bounded same-key replacement; a lost response is not proof no lease exists. A changed source or foreign/active process identity cannot inherit authority. Uncertain admission and missing executable return redacted unrun88; an actual nonzero child exit retains its code result. Windows pnpm shims use the shared command resolver. Reports of the original live claim latency remain reproduction leads; mock response-loss and native Windows launch fixtures prove these boundaries.

Windows host guard interpreters (BI-BA5F777A)

Windows may resolve bash to the WSL launcher while sh is absent. Native Windows fixture paths cannot be passed unchanged to WSL. Shell fixtures use scripts/lib/guard-bash.mjs to probe native path support in Git Bash discovered through the process PATH, a custom Git installation, or standard machine/user install locations. Host preflight, PR readiness and the policy runner share a child-only shell environment, so nested fixtures find Bash, sh and Git utilities consistently. Missing compatible Bash is a GuardRuntimeEnvironmentError, not a passing test or a product validation failure. Host PATH is unchanged. Git isolation fixtures write files through Node instead of invoking a shell. PKI rejection assertions and existing POSIX permission checks remain intact.

macOS Docker memory allocation (BI-DC4352E8)

An oversized Docker VM can compete with native applications and models even when container usage is modest. The installer now bounds local Desktop memory to a quarter of host RAM (8 GiB floor, 32 GiB cap), preserving valid smaller allocations. It verifies the running VM, backs up settings, restarts gracefully with consent for running containers or enabled Docker AI, and restores the previously running containers. Missing, malformed, symlinked or managed settings fail before shutdown. See the Mac guide. This is installer convergence, not an automatic migration of existing installs or proof of the cause of historical Chatterbox growth. Native model memory is outside the Docker VM allocation.

A running tally of platform-specific gotchas to watch for when supporting DPF’s target environments. This is the operational companion to the canonical design docs — it tracks symptoms and recurring traps, not contracts:

If a rule is universal across deployments, it belongs in the deployment contracts, not here. This file is for “when you touch X on platform Y, watch out for Z.”

Repository-local Codex endpoint overrides (BI-9F258707)

A trusted project’s .codex/config.toml overrides the user’s connector settings. Updating only the global file can leave the root checkout on HTTP while a new worktree uses HTTPS. Bootstrap now converges an existing project DPF block to its resolved endpoint; it does not create project overrides or scan other repos. Both repository updater wrappers pass the project explicitly to the standalone fallback. Disabled connections and custom credentials remain unchanged.

Malformed project TOML stops convergence. The standalone fallback validates it with Python 3.11’s tomllib, or tomli when already available on older Python; without either parser it refuses the write. The normal TypeScript planner uses its bundled TOML parser. Windows native and certificate-renewal acceptance need separate runtime evidence; passing Mac source tests does not establish them.

Target environments

Linux readiness evidence and remaining acceptance checks are maintained in the Linux guide. Publishing both CPU architectures does not certify either host’s full lifecycle. BI-3BE9A85C owns the Linux GA work; BI-E1AA1B3C repairs its Docker prerequisite path.

Environment Status LLM provider Host telemetry exporter Autostart
Windows 10/11 (Docker Desktop) GA Docker Model Runner windows_exporter on host (windows-host job, :9182) Scheduled Task
macOS (Apple Silicon, Docker Desktop 4.40+) Early access Docker Model Runner none (Docker Desktop VM hides host NICs) LaunchAgent
Linux (native Docker Engine) Early access Ollama (in-compose) node-exporter (linux-monitoring profile) systemd user unit
Cloud VM / TAPPaaS / Edge Node Spec-only (“design partner wanted”) per deployment per deployment per deployment

Out of scope (preflight refuses): Intel Mac, Windows-on-ARM, WSL2 without Docker Desktop, rootless Docker, Podman/containerd, distros older than Ubuntu 22.04 / Fedora 39 / Debian 12, air-gapped Linux.

How to use this file

For @parcel/watcher 2.6.0, retain optional dependencies: Windows x64, macOS arm64, and Linux x64/arm64 glibc/musl use locked prebuilt binaries. The canonical allowBuilds policy denies its opt-in node-gyp install hook. A target lacking a prebuilt binary needs a compatibility review, not a silent source-build fallback. A Windows snapshot/create-event check passed; cross-platform runtime verification belongs to the canonical build. See the bounded-build dependency disposition.

1. Monitoring & telemetry

# Symptom Platforms Root cause Status Watch for
M1 Platform Health shows CRITICAL ContainerDown / windows-host macOS, Linux Base prometheus.yml scrapes the Windows-only windows_exporter (host.docker.internal:9182), which can never come up off-Windows → ContainerDown fires. ✅ prometheus.macos.yml + prometheus.linux.yml drop the job; base keeps it for Windows. Locked by apps/web/components/monitoring/prometheus-config.test.ts. Any new host-OS-specific scrape target must be gated to the substrate whose overlay mounts it — never the shared base config.
M2 Monitoring summary reads “Degraded — Host telemetry is not available” on macOS macOS deriveMonitoringSummary (health-summary.ts) treated host telemetry as required; macOS intentionally ships neither windows-host nor node-exporter. ✅ Now distinguishes absent (job not a configured scrape target → expected on this substrate → Active) from down (up==0 → still Degraded). Locked by health-summary.test.ts. When adding a new host exporter, add its job to HOST_TELEMETRY_JOBS so its absence stays “expected” off-substrate.
M3 Grafana “Open dashboard” link goes to a dead URL on non-default / remote Grafana all (esp. macOS/Linux remote) SystemHealthDashboard.tsx:160 hardcodes href="http://localhost:3002". Roadmap Phase 3 claimed this was fixed; it was not. ⚠️ open One-line fix: href={process.env.NEXT_PUBLIC_GRAFANA_URL \|\| "http://localhost:3002"}. Touches .tsx → run next build.
M4 cadvisor / node-exporter crash-loop or refuse to start on macOS macOS They bind-mount /proc, /sys, /var/lib/docker, and rootfs /, which don’t exist / aren’t usable in the Docker Desktop Linux VM. 📌 Gated behind the linux-monitoring profile; only docker-compose.linux.yml opts in. Don’t add host-path bind mounts to any service started by the macOS or Windows overlay. CI compose-render asserts no /proc,/sys,/var/lib/docker mounts in the macOS rendered config.
M5 Local completion starts while a game or another app owns the GPU; decode collapses and Docker Model Runner stays resident Windows Docker Desktop The portal container cannot run nvidia-smi (no GPU device) and windows_exporter on :9182 publishes no GPU memory or utilization series. The in-process inference lock cannot see another process. Docker’s llama.cpp backend status can also report a missing nv-gpu-info.exe while the model still loads — that error is not “GPU busy”. ✅ scripts/publish-host-gpu.ps1 writes host-gpu.json into the state dir (DPF_HOST_GPU_SNAPSHOT_FILE overrides the portal path, default /dpf-state/host-gpu.json). Admission defers local completion on a fresh busy snapshot or an in-use runner, and unloads an idle runner when the card is busy. A missing snapshot does not defer. Do not scrape windows_exporter for VRAM, do not add a GPU bind mount, and do not treat the llama.cpp install error as a permanent deferral.
M6 Product health leads with Critical — Service adp is down and an ADP DOWN: … no such host tile while the same page lists adp as Optional — inactive all (first seen on macOS) prometheus.{linux,macos,dev}.yml scrape adp:8600 statically, but the adp service exists only while runtime:adp-integration is active, so every install without ADP carries a permanently down target that fires ContainerDown. ✅ BI-36DE938C: the health page checks target signals against the capability service projection. A target whose instance host is a service of an inactive capability (inactive-scrape-targets.ts) does not count toward Platform Status, raises no banner, and its tile reads Inactive (health-summary.ts). With the capability enabled, the same target still reports DOWN. Locked by health-summary.test.ts, inactive-scrape-targets.test.ts and ServiceHealthDashboard.test.tsx. A scrape target for a capability-gated service must address that service by its compose service name, so the instance host maps onto the capability catalog. The header health dot and the alert-to-inbox bridge do not load the capability projection yet.

Sweep landmine (do not trip): the discovery network sweep (packages/db/src/discovery-collectors/network.ts) depends on windows_net_nic_address_info (Windows) or node_network_info (Linux node-exporter) to enumerate real host NICs. The windows-host scrape must stay in the base prometheus.yml for Windows. Removing host exporters on macOS/Linux is safe only because the sweep already degrades to os.networkInterfaces() there (Docker Desktop VM boundary hides Mac/Windows host NICs regardless). See the “Network sweep data path” decision in the roadmap before touching any host exporter.

2. Shell scripts (BSD vs GNU coreutils)

Local-CI cleanup must never traverse a recycled PID tree. Revision 2 uses an owned Windows Job Object, process creation identity checks, and durable execution checkpoints. Parent-pipe loss closes the owned job; other process identities are never inferred from a matching command line. Windows loss and exact-handle tests cover the implementation. POSIX identity/signal races remain explicitly bounded in the recovery procedure. The admission floor retires older gate clients after the server is upgraded.

macOS ships BSD userland; Linux ships GNU. The installer targets bash 3.2 and python3 3.9 (stock macOS) — no associative arrays, mapfile, or ${var^^} in bash; no Python-3.10+-only stdlib kwargs or syntax in scripts the bootstrap runs with the system interpreter.

# Trap Platforms Status Watch for
S9 Engine-only Docker install reaches deployment without Compose; failed group/service setup looks successful Linux ✅ BI-E1AA1B3C: coherent official Docker package family, Compose preflight and explicit errors; behavioral tests in scripts/installer/lib/docker.test.mjs. Native clean-host runtime acceptance remains unrun. A Docker CLI version does not prove Compose exists. Capture status 75 in a conditional and explicitly handle failures inside the helper, because that conditional disables Bash errexit.
S1 sed -i differs (BSD requires a backup-suffix arg) macOS ✅ Use dpf_sed_inplace() in scripts/installer/lib/platform.sh — never raw sed -i. New scripts calling sed -i directly.
S2 netstat -anP tcp (-P is GNU-only) macOS 📌 Works today only because preflight.sh tries lsof → ss → netstat and macOS always has lsof. Don’t reorder the fallback chain or hardcode netstat -anP.
S3 readlink -f, stat -c, date -d, find -printf, grep -P macOS ⚠️ watch These GNU-isms have no BSD equivalent. Prefer POSIX forms; shellcheck --shell=bash runs in CI.
S4 Path.write_text(newline="\n") raises TypeError: unexpected keyword argument — the kwarg was added in Python 3.10 macOS (system python3 can be 3.9) ✅ packages/dpf-skill-pack/scripts/update_agent_toolchain.py writes bytes (write_bytes(content.encode("utf-8"))) to keep LF endings; crashed dpf-bootstrap-agent-toolchain.sh during ensure_codex_marketplace. Locked by update_agent_toolchain_test.py. Any Python invoked via bare python3 from installer/bootstrap scripts must run on 3.9: no match, no tomllib (3.11+, test-only), no 3.10+ stdlib kwargs. CI runs 3.11+, so a 3.9 break won’t show there — check docs.python.org “Changed in version” notes when touching these scripts.
S5 A synchronous Bash contract test crosses Vitest’s timeout, later reports status 127, or shifts the timeout to another case Windows contributors using Git Bash ✅ portal-migrate-boot.test.ts gives every real-script subprocess its own process-group deadline, captures command/status/signal/stdout/stderr, and retains a non-GNU watchdog fallback for macOS. Repeated --maxWorkers=4 coverage locks the boundary. Vitest’s timer cannot interrupt a blocked spawnSync. Any synchronous shell-contract test needs a subprocess-owned deadline shorter than the test timeout; do not assume GNU timeout exists on stock macOS.
S6 Local-CI FIFO remains frozen on dependency-convergence-active after the convergence owner exits Windows (also applicable cross-platform) ✅ Convergence locks now carry PID + process-start identity. Admission reconciles a dead/recycled owner atomically, keeps a live or unmeasurable owner fail-closed, and retains a bounded legacy-empty-lock compatibility path. Locked by local-convergence-lock.test.mjs and host-pressure acceptance. A directory’s existence or mtime is not owner liveness. Use the shared convergence-lock helper; never delete the lock manually or weaken active-install exclusion.
S7 Managed worktree bootstrap reports node_modules_missing even though pnpm is installed Windows ✅ Host command resolution now launches the pnpm cmd shim through ComSpec (without global shell mode), and bootstrap launch/install failures surface structured managed_install_failed evidence. Locked by host-command-invocation.test.mjs and bootstrap-worktree-deps.test.mjs. Node child-process APIs can return ENOENT for pnpm and EINVAL for direct pnpm.cmd execution. Reuse resolveHostCommandInvocation; do not add another executable-name fallback chain.
S8 Local-CI aborts convergence or rewrites pnpm policy when an agent host injects a newer pnpm than the repository pin macOS/Linux contributor hosts (observed with pnpm 11 over repository pnpm 10) ✅ The admitted runner checks package.json#packageManager; a mismatch is shadowed by a slot-local shim that invokes the exact repository pin through the available host pnpm. Matching hosts stay unchanged. Locked by local-ci-runner.test.mjs. Never trust contributor PATH as the package-manager version contract. Keep the repository pin in force for convergence, Prisma generation, tests, typecheck, and build.
S9 git, docker and prisma all fail with dpf-shell-guard.ps1: Parameter cannot be processed because the parameter name '' is ambiguous — for the whole user account, in every terminal, persisting across reboots; the installer itself then dies at “Step 7: Starting the platform” Windows ✅ The generated safety-bin\*.cmd shims no longer pass a -- separator (pwsh -File parses -- as a parameter prefix with an empty name), and dpf-shell-guard.ps1 is a simple script — no [Parameter()] attributes, no [CmdletBinding()], reading pass-through args from the automatic $args. Locked by check-shell-guard-shim-contract.test.mjs (policy guard shell-guard-shim-contract). Step 6 prepends safety-bin to the user PATH in the registry, so any binding bug there breaks the user’s real toolchain, not just DPF. Never make the guard an advanced script: common parameters reappear and forwarded flags prefix-match them (docker compose up -d → -Debug, docker compose -f → a -ForwardedArgs-style param). Never name a parameter $Args (collides with the automatic $args).
S10 A Node ESM entry script exits 0 silently — no output, main() never ran — when invoked through a symlinked path spelling; on macOS this broke the Janitor Tests bundle because os.tmpdir() is /var/folders/… → /private/var/…, so a contract test staging gate-worktree.mjs into tmpdir got a false “pass” instead of the refusal exit macOS (any host where the invocation path crosses a symlink) ✅ Node realpath-resolves the ESM entry module, so import.meta.url and process.argv[1] can spell the same file differently and the naive argv[1] === fileURLToPath(import.meta.url) guard fails open. The pregate script family (gate-worktree.mjs, pregate.mjs, pregate-status.mjs, local-ci-runner.mjs) now shares scripts/lib/entry-module.mjs (isEntryModule), which compares realpaths. Locked by entry-module.test.mjs and a symlinked-invocation test in tests/release/pregate-node-gate-contract.test.mjs. Never write a raw argv[1] === THIS_FILE guard in a new .mjs entry point — use isEntryModule(import.meta.url). For a gate script this failure mode is the worst kind: it fails open with exit 0. Linux CI can’t catch it (its tmpdir isn’t a symlink), so a green PR proves nothing about this trap.
S11 Host thrashes (load 20+, mds_stores / Search + Defender pegging CPU) while nearly idle — the OS file indexer grinds git-worktree sprawl (hundreds of worktrees, each a real node_modules tree = millions of files), starving the portal/inference VM so coworker dispatch, local-CI and the build gate time out / OOM macOS (Spotlight) and Windows (Search Indexer + Defender real-time AV) ✅ Every install surface — customer AND full-source, both OSes (install-dpf.sh, setup.sh, fresh-install.ps1, setup.ps1) — calls one shared helper (scripts/setup-worktree-hygiene.{sh,ps1}) that (a) registers the worktree janitor on a daily schedule (Tier A merged+clean only) so sprawl cannot accumulate, and (b) excludes the worktree base from the OS indexer: macOS adds it to the Spotlight Exclusions/Privacy list when it can elevate (interactive install → one sudo prompt, or already root) and prints the one-line step otherwise; Windows Add-MpPreference -ExclusionPath. The janitor + scheduler already existed but were never wired into any install surface. Locked by check-setup-worktree-hygiene.test.mjs. (BI-5F4F0146) A scheduler that exists but is never invoked at install is not a fix — wire it into ALL surfaces via one helper, not four copies. mdutil -i off does NOT work on a subfolder, and a plain .metadata_never_index proved insufficient on recent macOS (mds kept indexing) — the Privacy/Exclusions list is the mechanism that actually stops it, and it needs elevation (the elevated installer applies it; the unprivileged contributor bootstrap falls back to the printed step). Editing VolumeConfiguration.plist is SIP-fragile across macOS versions — validate per version. Prefer node_modules as junctions to the root clone; real per-worktree trees multiply the index/scan cost.
S12 Two local-ci-runner tests fail on every Windows host with local CI could not inspect host pnpm: unknown error, while staying green in CI Windows ✅ The fixtures wrote a #!/bin/sh fake pnpm, chmoded it 0755 and let preparePinnedPnpmEnvironment run the REAL spawnSync against it. Windows has no shebang support, so the spawn fails and the guard reports unknown error (a failed spawn leaves stdout and stderr empty). The tests already passed platform to fix the branch under test; they now also pass spawnSyncImpl, so the version probe is deterministic on every host. The one assertion that genuinely needs a POSIX shell — executing the generated shim — is its own test, skipped with a stated reason. A new test covers the platform: "win32" pnpm.cmd branch, which had no coverage at all. Locked by local-ci-runner.test.mjs. (BI-785B3253) Passing platform to a helper makes its DECISION deterministic, not its EXECUTION. If the helper takes a spawnSyncImpl (or any injected runner), a test that omits it is asserting against the host. Never let a fixture rely on a shebang, and never leave the platform branch your own host takes untested — that asymmetry is how this survived.

3. Docker / Compose

# Trap Platforms Status Watch for  
D1 Container cannot see the Mac/Windows host’s physical NICs macOS, Windows (Docker Desktop) 📌 Architectural limit of the Docker Desktop Linux VM — no network_mode: host / macvlan / CNI pierces it. Don’t promise host-LAN topology from in-VM containers on Docker Desktop. The Edge Node “Mode B” native helper is the long-term answer (see roadmap “Future direction”).  
D2 host.docker.internal unresolved Linux native Docker ✅ extra_hosts: ["host.docker.internal:host-gateway"] added to host-reaching services. Add the host-gateway entry to any new service that must reach the host on native Linux; it’s harmless on Docker Desktop.  
D3 Wrong LLM provider endpoint called Linux (Ollama) vs macOS/Windows (Model Runner) 📌 runtime:external-ai activates the profiled in-compose ollama service on Linux; docker-compose.linux.yml routes consumers to it without making portal startup depend on provider health. Entrypoints remain provider-aware (DPF_LLM_PROVIDER). Don’t call Model-Runner-only endpoints (/models/create) when provider is ollama, and never restore a portal hard dependency on an optional provider service.  
D4 Core runtime service hidden behind a profile all 📌 Profiles are for debug/test/monitoring only; core services must start by default. CI compose-render enforces this.  
D5 Self-upgrade “succeeds” but the running portal is a stale image (recreate didn’t swap) from-source installs (all OSes) ✅ BI-C8E90A79 defense-in-depth: the Dockerfile ALWAYS bakes /app/.dpf-source-content-hash from real source bytes (independent of the DPF_VERSION label), promote.sh adds a content-verify step asserting the running container’s hash equals the freshly-built image’s, and /api/platform/version surfaces sourceContentHash. Complements #1272’s source-truth stamp (promote.sh stamps rev-parse HEAD of the prepared source, not the requested target). A stale image left from a prior broken upgrade can carry the same SHA label, so SHA-verify alone can’t catch it — keep the content-hash swap guard whenever the promoter recreates the portal.  
D6 browser_profiles volume holds impersonation-capable session cookies (secret material, not ordinary evidence) all (EP-BROWSER-DRIVE) ⚠️ New volume on the browser-use sidecar (/profiles, spec §8.9). Sidecar is the only writer; the portal must never mount it (unlike browser_evidence). Backup/restore and uninstall flows must treat browser_profiles as secret: encrypt or exclude from ordinary evidence backups, and wipe on service-account de-provision. Don’t add a portal mount of this volume. Per-(account, site) user-data-dir isolation prevents cross-site cookie exposure.  
D7 Portal image build fails in Linux container with Turbopack duplicate-asset emission even though host next build passes Linux image builds (Node 24 Alpine) ✅ Portal image build uses Next’s supported webpack builder (pnpm --filter web exec next build --webpack) until the Linux Turbopack path is stable. Locked by scripts/lib/dockerfile-portal-build.test.mjs; validated with isolated compose image build. Do not reintroduce NEXT_TURBOPACK_USE_WORKER=0 for the portal image without rerunning the Linux image build. Host-only next build is not enough evidence for this trap.  
D8 Fresh install shows code graph as missing and scheduled-job status rows are empty macOS first-run observed; any compose install if env flags unset Base compose enabled Inngest self-sync but did not enable the scheduled function catalog or startup job registration, so code-graph-reconcile never registered or ran. ✅ Base compose now defaults DPF_SCHEDULED_INNGEST_FUNCTIONS_ENABLED=1 and DPF_OPTIONAL_STARTUP_TASKS_ENABLED=1 for installed runtimes. Keep library defaults conservative for tests, but compose installs must explicitly opt into cron/background job registration.
D9 Capability fixture renders on one host but fails when containers start on another Windows, macOS, Linux docker compose config validates topology but does not validate host-only bind mounts or device reservations. linux-monitoring remains an explicit Linux-host profile even when deep observability is enabled; GPU-backed tts likewise remains a compatibility profile with host prerequisites. 📌 Capability profile selection is host-aware; the source guard validates dependency closure, while installer harnesses must render on all three hosts and runtime verification must exercise the target host. Never infer runtime portability from a successful config --services render. Preserve linux-monitoring, tts, promote, dev, and integration-test semantics when adding capability profiles.
D10 A container-local path exported by the promoter overrides the install .env during docker compose up, remounting lifecycle state from /dpf-state on the host Windows and macOS Docker Desktop; any Compose host ✅ BI-91DAA63D: promoter internals use DPF_PROMOTER_STATE_DIR; DPF_STATE_DIR remains exclusively the install .env host interpolation variable across portal/sandbox recreation. Never pass DPF_STATE_DIR=/dpf-state into the promoter container. Contract tests must cover both promoter launch and promoter-owned Compose recreate boundaries.  
D11 Nearby DPF instances do not appear even though both portals are healthy Windows and macOS Docker Desktop; segmented enterprise LANs Containers cannot bind the physical host’s multicast interfaces, host firewalls may block UDP 5353, and mDNS is link-local by design. The first native implementation also passed the fully qualified _dpf-federation._tcp.local. name into zeroconf.NewType, whose contract requires only _dpf-federation._tcp; announcements then failed as dns: bad rdata while health was reported as ready. ⚠️ The native type-token defect is fixed and regression-covered; a macOS Bonjour browse sees the corrected advertisement. Host-native installer allocation landed, but Windows Scheduled Task install and real two-host add/remove remain open. Release assets are checksum-bound. Cross-VLAN and Internet links use invitation or a future governed discovery proxy. Never add mDNS to the base Compose topology or infer ownership/trust from a service record. Keep the library input type separate from the operator-facing FQDN. Verify Windows install, add/remove behavior on both hosts, UDP 5353 firewall behavior, and TXT privacy before closing V-01.
D12 Nearby discovery works but automatic pairing always reports tls_required Windows, macOS, Linux private-LAN installs A LAN IP or .local hostname cannot obtain public Web-PKI trust automatically, and the existing self-signed helper requires manual CA copying/mounting. Dockerized Caddy also cannot install its root into the host or peer portal trust stores. ⚠️ Organization Step CA selected by DI-236363AB3AA3; v0.30.2 is conditionally approved and the authority/leaf bootstrap substrate is implemented. Guided join-package UX, installer trust wiring, and real macOS/Windows acceptance remain open in BI-52D34506. Do not disable TLS verification, send invitations over HTTP, distribute a root private key, or treat a discovered certificate as trusted. Bind the CA to a private interface, pin its public-root fingerprint through the guided join flow, and retain dual approval.
D13 Login form looks inert on a valid password — post-login (or any Host-derived) redirect points at http://0.0.0.0:3000, which the browser cannot load, and no error is shown all (the portal binds 0.0.0.0 in-container; PUBLIC_URL/AUTH_URL unset on a default local install) ✅ BI-86165533 (#3380): with trustHost and no AUTH_URL, Auth.js can derive its base URL from the 0.0.0.0 bind address, so redirectTo="/workspace" resolved to http://0.0.0.0:3000/workspace. normalizeAuthRedirect (apps/web/lib/govern/auth-redirect.ts) rewrites a non-routable bind host (0.0.0.0, ::) to the operator-visible host (PUBLIC_URL when set, else loopback), wired into the Auth.js redirect callback while preserving the open-redirect guard. Never build a browser-facing redirect straight from the raw Host/x-forwarded-host header without normalizing the bind-all wildcard; set PUBLIC_URL when a real external host exists.  
D14 Edge action polling is configured on one host but silently absent after restart or unreachable from its peer Windows, macOS, Linux private-LAN installs The action path spans host-owned certificate/key files, a native Edge process, a persisted Compose secret overlay, private DNS, and TCP 8443; Docker Desktop cannot repair a missing host file or firewall rule. ⚠️ The trust bundle is now all-or-none, normal start/install paths restore docker-compose.edge-actions.yml, and Caddy provides a dedicated mTLS listener. Physical Mac/Windows reachability and lifecycle acceptance remain in BI-05EB708F. Never advertise action.execute from a partial trust bundle. Verify private-name resolution, TCP 8443, certificate renewal, quarantine, and revocation on each supported host; do not fall back to bearer-only HTTP.
D15 Candidate promoter preparation silently falls back to Docker’s deprecated legacy builder and a progressing cold-cache upgrade is killed after five minutes Windows, macOS, Linux Docker installs ✅ BI-2B4EAFC7: the portal runtime carries docker-cli-buildx, candidate and JIT promoter builds invoke docker buildx build --load explicitly, and candidate Docker operations use the governed DPF_PROMOTER_TIMEOUT_MS budget (25 minutes by default). Never rely on Docker’s legacy-builder fallback or give candidate preparation a shorter undocumented wall clock than the governed promoter budget. Retain cold-cache contract coverage and --load, because immutable digest inspection requires the built image in the local image store.  
D16 Background-job registration reports fetch failed and an admitted self-upgrade never dispatches on a source-free install Linux portal containers on all supported hosts ✅ BI-3FD07259: with APP_URL unset, Node resolved localhost to ::1 while the Next listener was reachable only on IPv4; the dispatcher then tried Git target discovery that a release-artifact install does not have. Self-registration now defaults to the server-owned 127.0.0.1 listener, persists only the real PUT outcome, and reconciles a fingerprint-valid stored release binding without creating another run. A container’s localhost address family is not the listener contract. For server-to-self calls, use the known listener address; never let a null source resolver override an already authenticated release admission or authorize a Git-source fallback.  
D17 Host admission undercounts reclaimable memory; local-CI can fence a GREEN image export or queue work despite available capacity macOS, Linux, Windows (Docker Desktop, WSL2) ✅ scripts/lib/host-available-memory.mjs owns Darwin reclaimable pages, Linux MemAvailable, and the docker-desktop guest cache (MemAvailable - MemFree) added to host free memory on win32 (BI-E129F788; Windows twin of BI-EB6DBAF0). The residual generic runner source repair now reuses that reader under BI-EB6DBAF0: its real-entrypoint fixture went RED at 4 GiB versus 42 GiB expected, then GREEN after the caller change; failed-probe fallback stayed 4 GiB. ⚠️ Canonical delivery and a fresh normally governed runner/scheduler acceptance remain pending; see the existing resilience design. Reuse the existing available-memory reader for every host claim; test the real claim payload and failed-probe fallback. Preserve profiles, reserves, queue order, leases and failure behavior. A fixture reading is not actual host capacity or runtime acceptance.  
D18 Office documents do not convert, and .docx, .xlsx and .pdf are not read: convertDocument returns converter-unavailable any host or target without a docker socket (cloud VM / TAPPaaS / edge shapes that do not mount /var/run/docker.sock), or an install with no pinned doctoolsImage 📌 By design (BI-52E565DA). The portal runs dpf-doctools one-shot through the docker socket, like the promoter; with no socket or no pinned image (name@sha256:), getConverterAvailability() reports no-docker-socket / not-configured, ingestion degrades to the S0 typed-unsupported result, a saved office document or PDF gets no renditions (one converter-unavailable lifecycle event per version; a bounded rendition backfill, drained in passes of at most 100 with each pass’s counts logged, is requested at portal start and on every pin change, debounced, BI-153EC72C, and also when the probe flips to available, which only fires where something scrapes /api/metrics, BI-9D43CBEF; a PDF owes only its plain_text rendition, BI-26CD1D1E), renderDocument (document generation, BI-3A0E5413) returns the same converter-unavailable without spawning, and dpf_dependency_up{service="doctools"} stays unset. Since BI-D1B40D43 (S9) the engine is also the only reader of .docx, .xlsx and .pdf: mammoth, read-excel-file and pdf-parse are gone and on the sbom deny list. On such a target a file upload or onboarding capture of one of them is stored with a typed unsupported result (word-document, workbook, pdf) whose reason names the file type and the fallback (plain text or CSV), a Workbooks or incumbent sheet import of an .xlsx is refused with the same reason, and brand-document extraction records an upload-pdf-no-text / upload-docx-no-text gap. CSV, TSV and plain-text files are read in-process and keep working, and the grid’s data-only .xlsx export (grid-xlsx.ts) still writes files. A tag-only reference is refused (image-unpinned). The image resolves in this order: the self_upgrade row’s doctoolsImage, then DPF_DOCTOOLS_IMAGE (source/dev), then, on a release install, the pin the portal resolves itself (BI-9A2EC54A). It gets that pin the way the promoter gets its own: at boot and every 20 min, ghcr.io/<GHCR_OWNER>/dpf-doctools:<immutable DPF_IMAGE_TAG> is resolved (the tag the running container was started from, not install-state’s, which promote.sh commits only after the new portal is verified; every boot logs a [doctools-image] boot: line, BI-903D22D0) with docker buildx imagetools inspect, the same digest publish-image.yml records. The pin is stored in self_upgrade.doctoolsImage and then pulled by digest, and availability reports image-missing until the pull lands. That window is closed ahead of time (BI-698B7F9A): a self-upgrade pulls the TARGET release’s dpf-doctools:<tag> in the candidate preflight, before the promoter and before the drain, and a pull that does not land fails the run as doctools-prepull-failed with nothing installed; the Windows consumer and Linux/macOS customer installers pull the same tag after the other release images and only warn if it fails. A release with no dpf-doctools manifest leaves the key unset and never blocks an upgrade or install, and a moving latest tag is not resolved. A customizable (source-built) install has no immutable DPF_IMAGE_TAG, so it resolves the same published image for the release tag its clone descends from, read from the git describe baked into the portal as DPF_PLATFORM_VERSION; a source upgrade re-stamps it, so the pin follows the upgrade (BI-4E18BC28). When no published image is reachable (offline, air-gapped, a lineage that predates dpf-doctools), it builds Dockerfile.doctools from a scratch copy of the clone’s two build inputs and pins the local image id, then retries the published image on every tick. Its source upgrades pre-pull too: the candidate preflight describes the TARGET commit (git describe --tags --always <target>) and pulls that release’s dpf-doctools:<tag> before the drain, but a pull that does not land only warns, because the install can still build the image after the swap. Compose stays free of any doctools reference (scripts/doctools-image-contract.test.mjs, AC-ODC-003), so the pin lives only in the portal’s own state. Locked by apps/web/lib/documents/conversion/*.test.ts, apps/web/lib/self-upgrade/doctools-release-image.test.ts, apps/web/lib/self-upgrade/doctools-release-image.boot.test.ts, apps/web/lib/documents/document-engine-boot.test.ts, apps/web/lib/self-upgrade/preflight.test.ts, apps/web/lib/self-upgrade/doctools-source-lineage.test.ts and scripts/installer/doctools-prepull.test.mjs. Never “fix” this by bind-mounting host paths or adding the engine to compose (an optional pinned image there froze promote-latest, PR #5290); input and output travel over stdin/stdout only.  
D21 An upgrade or build dies mid-way and every container shows the same start time macOS and Windows (Docker Desktop) Docker Desktop installs an update of its own and restarts the engine. Live 2026-10-02: eventInstallingUpdate at 14:28:41Z in com.docker.backend.log, every container (Postgres included) restarted at 14:29:10, and self-upgrade SUR-8782FCBD died mid-promotion. ✅ BI-75ECED42: the boot reconciler compares the database server start time with the run start and records engine-restarted (“The computer restarted Docker during the update”, retryable) instead of “orchestrator did not complete the swap”; scripts/setup.sh (macOS) and scripts/setup.ps1 advise turning automatic updates off. Turn off Docker Desktop automatic updates on an operated install, and update Docker deliberately between upgrades. When diagnosing, compare pg_postmaster_start_time() with the failed run’s start before blaming the promoter.
D19 A spatial view shows “region pack missing” even though a pack was copied onto the host all 📌 By design (BI-814F86E1, DI-63D94E36B0B0). Map packs live in the map_data named volume mounted at /var/lib/dpf/maps on the portal (override with DPF_MAP_DATA_DIR); the image never carries one. A pack copied to a host directory is invisible until it is placed in that volume with its <packId>.manifest.json, and a manifest whose byteLength disagrees with the file is reported as invalid (409). Never bind-mount a host map directory into the shared base compose file, and never fetch tiles from a remote host as a fallback.  
D20 Maps stop drawing after a Content-Security-Policy is introduced all ⚠️ open (watch). DPF has no app-wide CSP today. MapLibre 6 builds its worker from the same-origin module /api/map-assets/runtime/maplibre-gl-worker.mjs (BI-814F86E1), so any future CSP must allow worker-src 'self' blob:. When adding a CSP, include the worker directive and exercise a geographic view in light and dark themes.  
D23 Coworker reviews end without a receipt (spawn docker EAGAIN, “Sandbox file write failed”) and docker exec fails with “resource temporarily unavailable”; docker stats shows a container with tens of thousands of PIDs all Docker hosts ✅ BI-95BB9CB1: the sandbox’s PID 1 is a shell entrypoint and portal-tls’s is caddy; neither reaps orphaned children, so every git a build ran and every health-check probe left a zombie (75,868 and 11,573 measured on one install) until the Docker VM’s process table was nearly full. Both services now set init: true with a pids_limit (#6020), and scripts/compose-init-reaping.test.mjs keeps the init in place. Any new long-running service whose PID 1 is a shell script or an app that does not reap children (most are not init systems) must set init: true. Check docker exec <c> sh -c 'ps -eo stat' for Z states when a container’s PID count climbs.  
D24 An agent asks the operator a platform decision it could have put to the kernel (WWMD), or runs a command a guard should refuse; no PreToolUse guard blocks anything Claude desktop app (Code tab, Claude Code 2.1.295 / app 2.31226.1), all OSes and permission modes ✅ BI-E5C45F61 (2026-10-13). The host runs every PreToolUse hook, but it validates the output and accepts a top-level decision only as approve/block. The guards’ shared deny carried Grok’s decision: "deny" beside Claude’s hookSpecificOutput, so the host discarded the whole output, logged a hook_non_blocking_error in the transcript, and ran the tool. Claude and Codex payloads now get hookSpecificOutput only (hooks/lib/hook-io.mjs, contract in hooks/lib/claude-hook-contract.mjs); plugin 0.2.9. Bypass mode, an unresolved ${CLAUDE_PROJECT_DIR} and stale plugin copies were ruled out from the same transcript. The 2026-10-07 reading (BI-271B8EEC, “the app does not invoke the hook for AskUserQuestion”) was wrong; see the decision-routing spec §15.1. Probes, all harmless: docker compose up --help (compose-guard), npx tsc --noEmit (raw-tool-guard), npx prisma migrate dev with no DATABASE_URL (lease-punt-guard), a two-option architecture AskUserQuestion naming a BI id (decision-routing-guard). Each must be refused. SessionStart prints “PreToolUse guards inactive in this session” when a probe is not denied in the Claude contract or a declared guard command cannot run (hooks/pretooluse-guard-selftest.mjs). The transcript records a discarded verdict as hook_non_blocking_error. Client hooks only accelerate: the client-independent backstop is BI-271B8EEC Direction 2 (a decision reaches a person only with a qualifying kernel record), not built yet.  

4. Git / repo hygiene

# Trap Platforms Status Watch for
G1 git checkout/worktree add fails with “remote end hung up unexpectedly” macOS/Linux contributor without git-lfs ⚠️ The repo declares LFS (.gitattributes) but the LFS hook fails when the git-lfs CLI isn’t installed. Contributors: brew install git-lfs && git lfs install. To unblock a single op: GIT_LFS_SKIP_SMUDGE=1 git -c core.hooksPath=/dev/null <cmd>. Belongs in CONTRIBUTING.md.
G2 CRLF line endings break shell scripts in Linux containers Windows authors 📌 .gitattributes enforces eol=lf on *.sh and docker-entrypoint.sh. Don’t add .sh without LF enforcement.
G3 Multiple agent sessions sharing one working tree → branch/HEAD collisions, files swept into the wrong commit all ⚠️ recurring One session = one git worktree (AGENTS.md §4). Verify git worktree list and your current branch before committing when other sessions may be active.
G4 Committed .claude/settings.json hardcoded to a Windows host (powershell/pwsh hooks with absolute D:\DPF\… paths; //d/DPF/**, //d/backups//**, MSYS_NO_PATHCONV=… permission entries) → hooks silently no-op and entries are dead on macOS/Linux macOS, Linux contributors ✅ Hooks route through the pure-Node launcher scripts/hooks/run-hook.mjs (Node is the only guaranteed cross-platform hook runtime — the default hook shell is PowerShell on Windows, sh -c on Unix, and Git Bash is not bundled), which dispatches to the OS-appropriate scripts/<name>.{ps1,sh} via ${CLAUDE_PROJECT_DIR}. Per-machine paths/hooks moved to gitignored settings.local.json (template: .claude/settings.local.json.example). Don’t put absolute host paths, D:\…, or a shell:-pinned PowerShell/bash command in committed settings.json. New cross-platform hooks call run-hook.mjs <base>; machine-specific grants go in settings.local.json.
G5 Any pre-commit touching a doc with a ```mermaid fence dies with an execFileSync object dump — pid: 0, status: null, no stderr Windows ✅ render-doc-diagrams.mjs now resolves @mermaid-js/mermaid-cli/src/cli.js and runs it through the current Node, which needs no shell on any platform. It previously defaulted to node_modules/.bin/mmdc, a POSIX sh script Windows cannot exec (the runnable sibling is mmdc.CMD). A missing puppeteer browser now names npx puppeteer browsers install chrome-headless-shell instead of dumping the spawn object. (BI-334CB7DE) Never spawn a node_modules/.bin/<x> entry directly from Node — .bin holds an sh script plus a .CMD, and execFileSync can run neither without a shell. Resolve the package’s JS entry and run it through process.execPath.
G7 A test or script that reads a file next to itself fails with ENOENT on Windows only — CI is green, every open branch fails the local gate on the host Windows ✅ new URL(..., import.meta.url).pathname is /D:/repo/... on Windows, not a filesystem path; all 23 sites now use fileURLToPath. The no-url-pathname-fs-guard source guard (scripts/check-no-url-pathname-fs.mjs) refuses any new use outside a comment. Third recurrence: #4736, workroom-stall, #5247. (BI-5CC4159D) Derive filesystem paths from import.meta.url only through fileURLToPath from node:url. The .pathname.replace(/^\/([A-Za-z]:)/, "$1") workaround is the same defect and the guard flags it too.
G8 A local-CI gate is abandoned as blocked_control_plane_starvation docker:invalid-response while the host has plenty of free memory — the Docker engine is wedged (containers API answers HTTP 500, or the named pipe is gone), not starved Windows (Docker Desktop), any platform whose engine can die abnormally ✅ commandHealthy returned a bare boolean with stdio: "ignore", so docker info’s own distinguishing stderr was discarded and a wedged engine, an absent engine, a missing binary and a deadline all reported invalid-response. classifyDockerProbeFailure now answers engine-error (<status>) / engine-unreachable / binary-missing / timeout, and an engine reason adds an advisory that this is NOT host memory pressure. The wedge itself is BI-DDA569D9 (seven occurrences, each fixed by hand). (BI-3A008EBC) Two different remedies hide behind one symptom: drop the WSL page cache for real host pressure, move %LOCALAPPDATA%\Docker\run and %LOCALAPPDATA%\docker-secrets-engine aside and restart Docker for a wedged engine. Read the probe reason, not the word “starvation”. Never resolve a health probe to a boolean: the exit code and stderr are the diagnosis.
G9 A terminal window pops up every few seconds and steals keyboard focus while local-CI gates are queued Windows (Windows Terminal as default terminal) ✅ The durable-wait resumer is detached, so it has no console, and it started each gate re-claim without windowsHide. Windows gave every attempt a new visible console (BI-69178E02, #5502). The gate client runs from the branch being gated, so the fix did not reach branches cut before it. The claim server now retires those branches: a gate claim from a Windows worktree below GATE_CLIENT_REVISION 1 is refused with gate_client_upgrade_required (“rebase onto main”), which also stops their resumers. Any background process spawned detached must give every console child windowsHide: true. A client-side defect in branch-resident tooling is retired by raising the server’s floor in apps/web/lib/nonprod/gate-client-revision.ts, never by a per-host workaround.
G6 A doc’s mermaid diagrams are reported stale and re-rendered although no diagram changed; git diff shows nothing to explain it Windows authors, or any tool that rewrites a file with CRLF ✅ render-doc-diagrams.mjs strips CR before hashing fence content. Line endings are not part of a diagram’s identity. (BI-334CB7DE) A generator that content-hashes source text must normalise line endings FIRST. Git normalises on commit, so a CRLF working copy looks clean in git diff while every hash in the file has changed — observed as 466 CRLF pairs flagging all 4 fences in one page.

Room route guard on Windows

The room-addressing guard normalizes filesystem paths to forward slashes before walking the route tree and comparing baseline identities. Windows backslashes previously made existing route directories invisible and reported four valid baseline entries as stale. The Windows-path fixture and live-tree guard test cover this behavior; keep filesystem spelling out of portable route identities.

Recurring meta-pattern

Most entries above share one root: a Windows-first assumption baked into a shared/base artifact, inherited unchanged by macOS/Linux. G5 and G6 are the INVERSE — a POSIX-first assumption in shared tooling that breaks on Windows — so the question cuts both ways: a node_modules/.bin/<x> path, a sh-ism, or a hash over raw file bytes each assume a platform just as surely as a bind mount does. When adding anything host-coupled (a scrape target, a service, a bind mount, a host path, a default URL/port, a shell builtin), ask: “does this assume Windows/GNU/Docker-Desktop, and which substrate overlay should own it?” Put substrate-specific deltas in the owning overlay (docker-compose.{macos,linux}.yml, prometheus.{macos,linux}.yml), never in the shared base.

Persisted MCP endpoint recovery (BI-9F258707)

A bootstrap launched from a worktree may have no install PUBLIC_URL and may inherit a process environment older than setup. The existing client-environment resolver now reads the persisted user endpoint after the canonical install origin and explicit process override. POSIX decodes only managed export values as data; Windows reads the User environment. The saved Node CA bundle is considered before the default PKI file. OAuth credentials, grants, trust stores and unrelated exports are unchanged. Re-run the supported bootstrap to converge native client configuration; verify its effective endpoint, not only the saved environment. Tests cover POSIX recovery and precedence; native Windows execution must be verified on a Windows host.

Required services after container-driven upgrades (BI-FFFEA4ED)

Host Failure to watch for Recovery contract
Docker Desktop and native Docker A sibling promoter resolves relative monitoring binds beneath its own /host-source; the daemon cannot use that container path. A failed start leaves a Created container that an existence-only check misses. Translate binds using the promoter’s inspected host mounts. After release identity commit, release assets use the canonical install root. Retry only missing or proven never-started containers and retain failed recovery in the durable outcome.
Apple Silicon macOS The capability catalog asks for Docker speech while the macOS overlay points at the native service on port 8771. The canonical container catalog excludes macOS for dpf-tts; verify native synthesis separately. A disabled local.dpf-chatterbox-tts LaunchAgent must be restored through the existing native setup procedure when speech is requested.
All sandbox hosts The persistent source volume contains newer package manifests but older dependency links. Next fails before its health endpoint starts. The image-baked sandbox entrypoint performs a frozen dependency install and Prisma generation before serving the existing source. A failed install cannot fall through to a falsely ready preview.

Linux pilot startup repairs — 2026-10-03

PR #5998 merged the Docker Engine/Compose prerequisite repair. Additional source repairs prepare the runtime signing key and host-owned federation directories (BI-DFF7346E), remove the PKI host-UID assumption and suppress first-init password output (BI-F3AD783E), and pin pnpm plus Ollama loopback exposure (BI-5282E468). The Linux guide owns release and acceptance caveats. Confirm the selected release contains the changes; customer HTTPS/login, AI, reboot, restore, upgrade and uninstall remain unrun. Historical root-owned directories are preserved and refused for scoped recovery.

Edit this page on GitHub