# Deployment Audience: the IT admin standing this up inside the firewall. This covers the **SQL-backed deployment** — NGINX serving the static front end and a Python API backed by **PostgreSQL**. The repo already contains everything needed to run it as a Docker stack: `Dockerfile`, `docker-compose.yml`, the `nginx/` config, the front end in `html/`, and the API in `server/`. The detailed container reference (endpoints, password rotation, day-to-day commands) lives in [`server/README.md`](server/README.md) — this doc is the start-to-finish guide. ``` [ your TLS reverse proxy / traefik ] ← HTTPS terminates here │ (external "proxy" network) ┌────▼────┐ internal network ┌──────────┐ ┌────────────┐ browser ───────────────────────│ nginx │ ───── /api/ ───────> │ api │ → │ postgres │ │ (html/) │ │ FastAPI │ │ (db) │ └─────────┘ └──────────┘ └────────────┘ ``` Everything runs inside your firewall; the app makes **no outbound internet calls** (logo and scripts are local). > **Architecture note:** all static files live under **`html/`** and are *baked > into the nginx image* at build time (not bind-mounted). So after any front-end > change you rebuild the `webserver` image (see *Updating* below). The API image > is built from the root `Dockerfile`. --- ## 1. Prerequisites - A Linux host with **Docker** and **Docker Compose v2** (`docker compose …`). - An external Docker network named `proxy` that your TLS-terminating reverse proxy also sits on (the compose file marks it `external: true`): ```bash docker network create proxy ``` If you don't run a separate reverse proxy, you can instead publish the nginx container's port 80 directly (see the note in step 4) and terminate TLS there. - The repository checked out on the host. ## 2. Create the database credentials (`.env`) Create a file named `.env` in the **project root** (same folder as `docker-compose.yml`). It is git-ignored and must never be committed. ```bash # .env — project root POSTGRES_DB=wpsuite POSTGRES_USER=wpsuite POSTGRES_PASSWORD= # REQUIRED — signs login session cookies, AFTER Okta has confirmed who someone # is. If unset, `docker compose up` errors out and the API refuses to start. # Generate once and keep it stable: # openssl rand -base64 48 AUTH_SECRET_KEY= # REQUIRED (in spirit — see the note below) — Okta OIDC is the only sign-in # path (D15/D16). There is no local password anywhere in this app to fall back # to, so without these nobody can sign in at all. Get them from the Okta app # integration (sign-in method OIDC - Authorization Code, Web Application): # # OKTA_ISSUER the authorization server, e.g. # https://your-org.okta.com/oauth2/default # OKTA_CLIENT_ID \_ from the app integration # OKTA_CLIENT_SECRET / # OKTA_REDIRECT_URI must exactly match a "Sign-in redirect URI" # registered on the app integration, e.g. # https://wp-suite.company.local/api/auth/okta/callback # # Full explanation, and the optional OKTA_IDENTITY_CLAIM override, in # server/.env.example. Not enforced at startup the way AUTH_SECRET_KEY is — # the API starts without these, it just refuses every sign-in and says so in # `docker compose logs api` (server/okta_auth.py's describe()). OKTA_ISSUER= OKTA_CLIENT_ID= OKTA_CLIENT_SECRET= OKTA_REDIRECT_URI= # Encrypts database backups at rest (AES-256). Set this BEFORE the DB holds # customer IP. Keep the passphrase OFF this host — losing it makes dumps # unrecoverable: openssl rand -base64 32 BACKUP_ENC_PASSPHRASE= # OPTIONAL — SMTP password for WP-assignment email. Email is OFF by default and # enabled from the Admin console; the host/port/from-address are configured # there, but the password is only ever read from this variable (never stored # in the DB or shown in the UI). Leave unset until you have SMTP details. # SMTP_PASSWORD= ``` The API builds its own DB connection string from the `POSTGRES_*` values and **encodes the password automatically**, so a password with special characters (`@ ! # : /` …) works without any manual escaping. `DATABASE_URL` is **optional** and only needed if you want to point the API at some other database; if you do set it, you must URL-encode the password yourself, and it's ignored whenever the three `POSTGRES_*` values are present. Generate a strong password with `openssl rand -base64 32`. > **Portainer note:** for a Git-based stack these go in the stack's > **Environment variables** section (Portainer doesn't read a local `.env`). > Set `POSTGRES_DB` / `POSTGRES_USER` / `POSTGRES_PASSWORD` / `AUTH_SECRET_KEY` / > `OKTA_ISSUER` / `OKTA_CLIENT_ID` / `OKTA_CLIENT_SECRET` / `OKTA_REDIRECT_URI` / > `BACKUP_ENC_PASSPHRASE` (and `SMTP_PASSWORD`, if you enable email) there. These are the only credentials in the system, and they never appear in the compose file or in git. ## 3. Point your reverse proxy at the nginx container The nginx container listens on port **80** on the `proxy` network and expects TLS to be terminated upstream (by your reverse proxy / traefik). Route your chosen hostname (e.g. `wp-suite.company.local`) to the `nginx_webserver` container on that network. The container already proxies `/api/` to the `api` service internally — no extra app config needed. > **Serve it over HTTPS, and forward the scheme.** The bundled nginx sets the > security response headers (CSP, HSTS, `X-Frame-Options`, `nosniff`) and passes > `X-Forwarded-Proto: https` to the API, which is what makes the session cookie > `Secure`. If you front the stack with your **own** proxy instead, make sure it > terminates TLS and forwards `X-Forwarded-Proto: https` — otherwise the login > cookie won't get the `Secure` flag. HSTS also assumes the site is only ever > reached over HTTPS. ## 4. Bring it up From the project root: ```bash docker compose up -d --build # builds the api + nginx images, starts all three containers docker compose ps # confirm nginx_webserver, wp_api, wp_db are running/healthy docker compose logs -f api # watch the API start (Ctrl-C to stop following) ``` The database schema is **created automatically** on first API start — no manual `CREATE TABLE`. The Postgres data lives in the named volume `pgdata` and survives `docker compose down` (only `down -v` deletes it). > No separate reverse proxy? Publish nginx directly by adding a `ports:` mapping > to the `webserver` service (e.g. `"8080:80"`) and terminate TLS at whatever > sits in front of it. The internal `api`/`db` containers should **never** be > published. ## 5. Verify ```bash # API liveness (from the host, through the proxy hostname) curl https://wp-suite.company.local/api/health # → {"ok": true} # Interactive API docs # https://wp-suite.company.local/api/docs ``` Then load the site in a browser: the home page should prompt to **select or create a project**. Create one, complete an SOP, and confirm a row appears: ```bash docker compose exec db psql -U wpsuite -d wpsuite -c "select id, name from projects;" ``` ### Automated smoke test `server/smoketest.py` exercises the whole stack end-to-end (health → sign-in → project → SOP → Work Package → the AWP issue gate → status → metrics → comments → archive round trip → cascade cleanup → sign-out). Stdlib only — no pip/jq. It **signs in first**, because every `/api/` route except `/api/health` requires a session — but there is no local password to sign in with (D15/D16), and Okta requires a real browser to complete, which this script cannot do. So instead of an HTTP login, it mints a session directly the same way `okta_callback()` does after Okta hands back an identity, which means **it has to run somewhere that can read the same `AUTH_SECRET_KEY` and reach the same database as the server under test** — inside the `api` container, or local dev against your own DB. It can no longer sign in to an arbitrary remote URL from an unrelated workstation the way the old password-based version could. The account named by `WP_SMOKE_USER` must **already exist** — sign it in through Okta once first, or pre-create it from the User Directory — and must be an **admin**: the run creates a project and deletes it again, and deleting one takes Project Admin on it. The script checks the signed-in role up front and warns if it is too low rather than letting you find out in the cleanup step. ```bash # From inside the api container — has AUTH_SECRET_KEY and DATABASE_URL, and # hits FastAPI directly. This is the normal way to run it in production: docker compose exec -e WP_SMOKE_USER api \ python /app/server/smoketest.py http://localhost:8000 # Local dev, against the app you're running yourself: export AUTH_SECRET_KEY=... DATABASE_URL=... WP_SMOKE_USER= python3 server/smoketest.py http://localhost:8000 # Add --keep to leave a demo project in the DB so you can open it in the UI. # --user overrides $WP_SMOKE_USER if you'd rather be explicit. ``` Exit codes: **0** all checks passed · **1** one or more checks failed · **2** the run could not start (host unreachable, or credentials missing or rejected). The last is kept separate on purpose — "I could not test this" is a different answer from "this is broken", and automation should not treat them alike. ### Front-end browser check `tests/browser_check.py` is the other half: the smoke test proves the API works, this proves the **pages** work. It runs them in headless Edge (or Chrome) over the DevTools Protocol and asserts what only a browser can settle — that each page boots without a JavaScript error, that the role-dependent renderings are right, and that the layout rules the console pages depend on are actually in effect. Self-contained: it creates a throwaway SQLite database, seeds a fixture (two projects, an admin, a Project Super User, a plain member, and accounts positioned to exercise in-scope / out-of-scope / invisible), starts its own server on a free port, and tears all of it down. **Your real database is never touched.** Stdlib only. ```bash python tests/browser_check.py # everything, ~71 checks python tests/browser_check.py --keep-server # leave it up to poke at by hand WP_BROWSER=/path/to/chrome python tests/browser_check.py ``` Same exit codes as the smoke test, including **2** for "no browser found" — a missing browser is not a failing app. Run this after any change to `html/users.js`, `html/wp-sidenav.js`, `html/console.css` or `html/admin.js`. It is the check that would have caught a rule lost while `console.css` was being extracted out of `admin.html`, which is a silent, whole-page regression that no server-side test can see. Exit code 0 and "ALL PASS" means the API, the Python logic, and SQL are all working. It cleans up after itself (the test project and its SOP/WPs are deleted via cascade); a single tagged test comment remains (there's no comment delete endpoint). ### Loadable demo project `server/seed_demo.py` populates a realistic **DEMO** project (a complete SOP plus a spread of Work Packages: issued, gated, a multi-discipline master with split instances, an overdue one, an over-threshold draft) so there's data to look at. ```bash python3 server/seed_demo.py https://wp-suite.company.local --insecure python3 server/seed_demo.py https://wp-suite.company.local --clean # remove it later ``` > **What shows where:** the DEMO **project**, its **SOP**, and its **Work > Packages** are all API/SQL-backed, so they appear in the home-page project > picker and render in the Creator/Dashboard as soon as any user opens the > project. Inspect them at the SQL layer with `smoketest.py` or: > ```bash > docker compose exec db psql -U wpsuite -d wpsuite \ > -c "select number, subject, status from work_packages order by number;" > ``` --- ## What is stored in SQL today The API + Postgres are the system of record. Everything below is server-stored and shared across every user who opens the project: | Data | Stored in PostgreSQL today? | |------|------------------------------| | **Projects** | **Yes** — the front end is API-first (`/api/projects`), falling back to the browser only if the API is unreachable. | | **Comments / feedback** | **Yes** — every feedback surface posts to `/api/feedback`. | | **SOPs** | **Yes** — pulled from `/api/sops` on load and written through on every save. | | **Work Packages** | **Yes** — same write-through to `/api/wps` (+ issue / status / archive / metrics), including the owner assignment (`assignee_id`). | Saves go through a **durable client-side sync outbox**: edits are written to the API immediately, and if the device is offline they queue and retry when it reconnects (4xx rejections are dropped rather than retried forever). The browser cache is only an offline fallback that reconciles through that outbox — so two users on the same project see the same server-stored SOP and Work Packages. ## Data model (PostgreSQL) | Table | Holds | Key columns | |-------|-------|-------------| | `projects` | top-level construction projects | `name`, `number`, `client`, `division`, `site`, `sample`, `archived_at`, `data` | | `sops` | project SOP baselines | `project_id` → projects, `name`, `number`, `complete`, `data` (full SOP JSON) | | `work_packages` | individual IWPs | `project_id` → projects, `sop_id` → sops, `parent_id` (split instances), `number`, `subject`, `type`, `status`, `assignee_id` (owner), `issued_at`, `archived_at`, `data` (full WP JSON) | | `comments` | feedback from any page | `source`, `sop_id`, `wp_id`, `step`, `author`, `text`, `extra` | | `users` | login accounts | `username` (matched against Okta's identity claim — no password column; D15/D16), `role`, `full_name`, `email`, `is_active`, `auto_add_projects` + `auto_add_role` (default membership on new projects), `token_version` | | `project_members` | per-project access control | `user_id` → users, `project_id` → projects | | `audit_log` | append-only activity trail | `actor`, `action`, `entity_type`, `entity_id`, `project_id`, `summary`, `detail` | | `notifications` | in-app record + email outbox | `user_id`, `kind`, `wp_id`, `subject`, `status` (pending / sent / failed / skipped) | | `app_settings` | admin-configured settings (e.g. email) | `key`, `value` (JSON) | The complete client document is stored verbatim in each row's `data` JSON column; frequently-listed fields are promoted to real columns for filtering. ### Endpoints (summary) Projects `GET/POST /api/projects`, `GET/DELETE /api/projects/{id}`, `POST /api/projects/{id}/archive` · SOPs `GET/POST /api/sops`, `GET /api/sops/latest`, `GET/DELETE /api/sops/{id}` · Work Packages `GET/POST /api/wps`, `GET/DELETE /api/wps/{id}`, `POST /api/wps/{id}/issue`, `POST /api/wps/{id}/status`, `POST /api/wps/{id}/archive`, `GET /api/wps/metrics` · Comments `POST /api/comments` (and `/api/feedback`), `GET /api/comments` · Auth `GET /api/auth/okta/login` / `okta/callback` (the Okta sign-in round trip), `POST /api/auth/logout`, `GET /api/auth/me`, admin user management under `/api/auth/users` (including `POST /api/auth/users/{id}/auto-add`) · Admin-only `GET/PUT /api/settings`, `POST /api/settings/test-email`, `GET /api/notifications`, `GET /api/projects/{id}/members`. List/latest/metrics accept a `project_id` (and `sop_id`) filter. `GET /api/projects` and `GET /api/wps` both take `archived=exclude|only|all` and **default to `exclude`** — anything that needs to see archived rows (the admin console, the demo cleanup) must ask for them. Full reference and request shapes: `/api/docs` and [`server/README.md`](server/README.md). --- ## Updating after a change ```bash git pull docker compose up -d --build webserver # front-end change (html/) — rebuild the baked image docker compose up -d --build api # backend change (server/) ``` ## Backups & retention A **`backup` sidecar** (in `docker-compose.yml`) runs `pg_dump` on a schedule and writes gzipped, timestamped dumps to `./backups/` on the host. It starts with the stack — no cron to set up. - **Cadence / retention:** daily, keeping the newest 14 dumps. Override in `.env` with `BACKUP_INTERVAL_SECONDS` (seconds between dumps) and `BACKUP_KEEP` (how many to keep). - **Encryption at rest:** set `BACKUP_ENC_PASSPHRASE` in `.env` and dumps are written AES-256-encrypted as `*.sql.gz.enc`. **Do this before any customer IP goes in** — without it the dumps (and every offsite copy) are plaintext. Store the passphrase somewhere other than this host; if you lose it the backups can't be restored. - **Ad-hoc backup now:** `docker compose exec backup sh /scripts/db-backup.sh` - **Restore (destructive — overwrites current data):** `docker compose exec backup sh /scripts/db-restore.sh /backups/wpsuite-YYYYMMDD-HHMMSSZ.sql.gz.enc` - **Offsite — do this:** the dumps live in `./backups/` on the host; if the host/volume dies, so do they. Sync that folder offsite from the **host** (e.g. a cron running `rclone`/`aws s3 sync`). The `db`/`backup` containers are on an egress-less `internal` network on purpose, so offsite must be pushed from the host. - **Test restores quarterly:** load the latest dump into a throwaway database and confirm it applies. An untested backup is not a backup. ## Field devices & data at rest The field view (PWA) caches a project's Work Packages/SOP in the browser's localStorage so it works offline — i.e. **customer IP sits on the device**. localStorage is not encrypted and is not a security boundary. Signing out clears the cached project data, but for any tablet/phone that opens customer-IP projects: - **Require full-disk encryption** (BitLocker / FileVault / Android FBE / iOS is encrypted by default) and a device passcode. - **Enrol field devices in MDM** so a lost device can be remotely wiped, and keep the browser profile per-user on shared devices. - Users should **sign out** when handing off a shared device (clears the cache). ## Email notifications (optional) Work-package **owner assignment** works out of the box (in-app only). Optional **email** on assignment is **OFF by default** and is turned on from the **Admin console → Notifications & email** card, where an admin sets the SMTP host / port / TLS / From address and flips the master toggle. - The **SMTP password is never stored in the database.** It is read only from the `SMTP_PASSWORD` environment variable (see the `.env` block in step 2 and the `api` service in `docker-compose.yml`). The UI shows only whether it is set. - Email stays effectively off until **all** of: the toggle is on, SMTP host + From are configured, and `SMTP_PASSWORD` is present. Until then, assignments are still recorded in-app (status `skipped`); nothing is sent. - Notification emails carry only a **WP number and a deep link** — never the work package contents — so customer IP stays behind the login. - Use the card's **Send test email** button to confirm SMTP before enabling. ### Sign-in and admin bootstrap (Okta) There is no local password anywhere in this app — no "Forgot password," no reset link, nothing email-related to sign-in (D15/D16). `SMTP_PASSWORD` above is purely for WP-assignment notification emails. Sign-in is entirely Okta's job: `login.html` redirects to Okta, and access control is **who is assigned to the app integration in Okta** — see step 2's `OKTA_*` variables and [`server/README.md`](server/README.md#sign-in-okta) for the full flow. The first admin has to sign in through Okta once (landing as an ordinary `project_user`, auto-provisioned), then get promoted from a shell: ```bash docker compose exec api python -m server.manage_users promote alice --role admin ``` This is deliberate, not an oversight: a hand-typed username at account-creation time risks a second, orphaned row if it doesn't exactly match what Okta sends, so the CLI promotes an existing Okta-provisioned row rather than creating one blind (D16). Every admin after the first can be promoted from the User Directory page — no shell access needed. **No break-glass path.** If Okta is unreachable or misconfigured, the app is unreachable for everyone, including admins, until Okta is restored — the same posture the abandoned LDAPS design took, carried forward deliberately (D16). ## Permissions roles `User.role` is the **permissions** role; `User.project_role` is the person's **job function** on the project (Project Manager, Superintendent, …) and grants nothing. Both are set on the **User Directory** page (`users.html`) — not the Admin Console, which no longer manages accounts. | Role | May do | |---|---| | `admin` | User administration everywhere, app settings, and every project | | `project_super_user` | Everything `project_admin` may do, **plus user administration on the projects they hold the role on**: pre-create accounts by username, set permissions, grant project access | | `project_admin` | On assigned projects: delete work packages, change a **completed** SOP, delete the project | | `project_user` | Create/edit work packages, author a SOP up to completion; may archive a WP but not delete one | Enforced server-side by `require_project_admin` in `server/app.py`; the front end only hides controls to avoid dead-end clicks. Accounts created before roles existed carried the role `user`, which the migration rewrites to `project_user`. ### Project Super User — what bounds it The role exists so a project admin can staff their own job without an app admin. Its limits are what make it safe to hand out, and all of them are server-side (`managed_project_ids`, `manage_user_problem`, `grantable_roles` in `server/app.py`): * **Scope comes from projects, not the job title.** A super user administers the users of the projects they hold the role on — via their account role, or via `ProjectMember.role` for a super user on one job only. No projects, no authority. * **Account changes need EXCLUSIVE scope.** Disabling, renaming, changing permissions or deleting are global acts, so they are refused when the target is also on a project the caller does not administer. The directory shows those rows read-only with the reason. An app admin has to make the change. * **No admin or super-user targets, and none granted.** A super user may hand out `project_admin` / `project_user` only, and may not touch an admin's or another super user's account — so the role cannot become a route to app-wide control. * **Saving project access never reaches outside scope.** `PUT /api/auth/users/{id}/projects` rebuilds only the caller's own slice; memberships on projects they don't administer are left untouched. * **App settings, feature flags and the default-member rule stay admin-only.** No migration is needed for the new role — `users.role` is already `String(20)` and `project_super_user` fits. Grant it from the User Directory (Permissions column), or per project from **Project access → Project Super User here**. ## Feature flags **Admin console → Features.** `bim_enabled` is **OFF by default**: the SOP creator hides the BIM/VDC section and every project is install-only (IWP). A SOP that already has BIM enabled keeps its data — it just stops being offered — so turning the flag off never deletes BIM types, gates, or sequence steps. ## Release gates (constraints + predecessors) A work package reaches **Issued** only when both gates are met: 1. every constraint is **Cleared** or **N/A** — a hard gate, no override; 2. every **predecessor work package** (`data.predecessors`, a list of WP ids) is **Closed**. Enforced by `enforce_release_gates()` on **every** path that can set a status — `/api/wps` (the browser and the offline outbox both save through it), `/api/wps/{id}/issue`, and `/api/wps/{id}/status`. Also: - **Overridable, deliberately.** Planners legitimately release ahead of upstream close-out, so the predecessor gate accepts `data.gateOverride = {reason, by, at}`. A blank reason is not an override. The server writes a `gate_overridden` audit event naming the reason and what was skipped, and the reason prints on the package. Changing the predecessor set clears the override. - **Cycles are refused** (`check_predecessor_cycle`) — direct and through a chain, with a 400 explaining which package already waits on this one. - **A deleted predecessor does not block.** It would otherwise freeze everything downstream of a package someone removed. - The Creator's picker hides itself and any package that already waits on it, so a cycle is hard to build in the first place; the dashboard refuses to issue a blocked package and points at the form for the logged override. `data.seq` (the SOP sequence phase) is still stored and shown, but it is descriptive — it gates nothing. ## Critical constraints reopened after release A constraint marked **Critical** on the SOP that reopens **after** the package was released emails the **owner, PM, CM and everyone on the package's distribution list** (minus whoever reopened it), and writes a `constraint_reopened` audit event. Detected by comparing incoming constraints against the stored ones inside the normal upsert — *not* a separate endpoint, because the browser saves through the sync outbox, which only replays `POST /api/wps`; anything hung off another route would be lost offline. It fires only on a real transition (cleared/N-A → open), so re-saving an already-open constraint doesn't re-announce, and never for a package that was never released or a non-critical constraint. Bodies carry the constraint name, WP number and a link — never the package contents. ## Localization (dates, times, numbers) Three levels, most specific first — resolved in `html/wp-format.js`: 1. **the user's own preference** — *Language & time* in the top-right menu (`users.locale` / `users.timezone`, via `POST /api/auth/preferences`) 2. **the app default** — Admin console → Features → *Localization defaults* (`default_locale` / `default_timezone`) 3. **the browser**, as before Timezone names are validated against the server's own `zoneinfo` database, and the picker is fed from `GET /api/timezones` so it can only offer what will be accepted. Calendar dates (a due date, a kitting date) are formatted from their parts and are **never** shifted by a timezone — only real instants (MIMO windows, history, notifications) are converted. Use the shared helpers (`wpFormatDate`, `wpFormatDateTime`, `wpFormatTime`, `wpFormatNumber`) rather than `toLocaleString()`, or a page will quietly ignore the preference. ## Top-bar chrome (project switcher + search) `html/wp-chrome.js` + `wp-chrome.css` inject a project switcher and a centered global search into whichever top bar a page has — the dark `.wp-appbar` or the older `.header`. It is skipped inside an iframe, so the embedded WP creator does not get a second bar. - Switching project reloads the current page with `?project=`; every page already resolves its project from that parameter. - Search calls `GET /api/search?q=`, which is **scoped to the caller's projects** (`scope_to_access`) and hides archived work packages, archived projects, and anything belonging to an archived project. LIKE wildcards in the query are escaped, so searching `100%` matches a literal `100%`. Two-character minimum. - Ctrl/Cmd-K focuses the field from anywhere. ## Schema migrations (Alembic) Schema is managed by **Alembic** (`server/alembic/`). The API container runs `alembic upgrade head` on startup (see the `Dockerfile` CMD), so **deploys apply pending migrations automatically**. - The **baseline** migration is idempotent: on a fresh database it creates every table; on a database whose tables already exist (made by the old `create_all`) it adopts the schema as-is — no manual `alembic stamp` needed. - Local dev on SQLite still auto-creates tables for a zero-config run; Postgres is migrations-only. - **To change the schema:** edit `server/models.py`, then generate and review a migration before committing: ```bash # from the project root (against your dev SQLite or a staging DB) python -m alembic -c server/alembic.ini revision --autogenerate -m "describe the change" python -m alembic -c server/alembic.ini upgrade head # apply locally to test ``` The next `docker compose up -d --build api` applies it in production on startup. ## Local trial without Postgres For a quick local look, the API falls back to a SQLite file when `DATABASE_URL` is unset (`sqlite:///./wpsuite.db`) — see [`server/README.md`](server/README.md) § *Local dev*. The front end alone can also be served statically from `html/` (it falls back to browser storage when the API isn't reachable). ## Per-project permissions `users.role` is the account's **default** permissions role. A membership row can override it **per project** (`project_members.role`), so someone can be Project Admin on one job and a plain Project User on another. Empty means "inherit the account's role", which is how every pre-existing membership behaves. Resolved by `effective_role()` in `server/app.py`; `require_project_admin()` uses it, so deleting a work package, changing a completed SOP and deleting a project are all judged **on that project**. An app `admin` is admin everywhere and bypasses membership entirely. Set it in **Admin console → User administration → Project access** (its own column, showing how many projects each account can reach). The dialog ticks project access and picks the role on each; `/api/auth/users/{id}/projects` takes `{project_ids: [...], roles: {project_id: role}}` and only accepts the two project-scoped roles. Changes are audit-logged as `project_access_changed`. **Who appears in the SOP's people pickers** is `GET /api/projects/{id}/members` — the project's members plus app admins, each with their effective role on that project. A project with nobody assigned shows only the admins, which is why assigning people is the first step on a new job. ### Default members on new projects Memberships are also created automatically. **Admin console → Default members on new projects** flags accounts (`users.auto_add_projects`) that belong on every job — the PM who runs them all, the QC lead — with the role they should hold there (`users.auto_add_role`, sharing `project_members.role`'s value space, `''` = inherit the account's own). - It applies **only to projects created after the flag is set**. Nothing is back-filled onto existing jobs; use **Project access** for those. - App admins are skipped (they already reach every project) and the flag is cleared if an account is promoted to admin. Inactive accounts are skipped. - Runs in `add_default_members()` on the `is_new` branch of `upsert_project`, so it covers every route into project creation — the home page, the sample project, the demo seeder. An update never re-runs it. - If the creator is themselves a flagged member, the membership created for them as creator carries their `auto_add_role`, so they aren't silently downgraded on the one job they started. - Audit-logged once per project as `project_access_granted` with `detail.reason = "auto_add_projects"`. ## Archiving a project A finished job is archived rather than deleted: `projects.archived_at`, set from **Admin console → Projects** (or `POST /api/projects/{id}/archive`, which needs Project Admin **on that project**, same bar as deleting it). An archived project is **hidden and frozen**: - It leaves the home picker, the app-bar switcher and global search, because `GET /api/projects` defaults to `archived=exclude`. - It is still readable by id, so a deep link renders it — with a read-only banner from `wp-chrome.js` — and the admin console still lists it under `?archived=all`. - Every write that lands on it is refused with **409** by `require_project_writable()`: saving a project, SOP or work package, deleting either, issuing, status changes, WP archiving, and comments on its WPs/SOPs. Moving a work package *into* or *out of* an archived project is refused too. 409 rather than 403 is deliberate — nobody lacks a permission, the project's state is the objection, and the browser outbox (`html/project-data.js`) retires 4xx ops instead of retrying them forever. - Unarchiving and **deleting** stay allowed: unarchive is the one write an archived project must accept, and archive-then-delete is a normal sequence. Nothing is removed, and unarchiving restores all of it. `server/smoketest.py` asserts the whole round trip. ## Asset freshness (why the app can't run half-updated) A page must never run against a stylesheet or script from a previous deploy. Three things enforce that, and all three are needed: 1. **`Cache-Control: no-cache` on HTML/CSS/JS** — set by NGINX (`nginx/conf.d/wp-suite.conf`) and by the dev server (`_NoCacheCode` in `server/app.py`). With no header at all the browser applies *heuristic* freshness, roughly 10% of each file's age, so the least recently changed file gets the longest lifetime — which is exactly how HTML and CSS drift apart. ETag/Last-Modified still make each revalidation a cheap 304. 2. **The service worker fetches code with `cache: 'no-cache'`** (`html/sw.js`) and precaches with `cache: 'reload'`. A plain `fetch(req)` inherits the request's default cache mode and consults the browser HTTP cache, so "network-first" alone was not enough. Non-`ok` responses fall back to the cache rather than replacing a page the cache could still serve, and cache keys drop the query string so in-app links (`?project=…&tab=…`) still resolve offline. 3. **Components whose CSS-missing state is *broken* carry their own critical layout.** The embedded creator's iframe keeps its sizing inline (and `sizeWPFrame()` re-applies it), and the work-package panel injects a floor of positioning rules from `wp-creation-app.js`. Both had failure modes — a 300×150 iframe, and panel controls dumped loose into the form — that a missing rule turned into a broken page rather than a plain one. If you change the shell file list in `sw.js`, bump `CACHE`. > **NGINX note:** the `Cache-Control` value comes from a `map $uri $wp_cache_control` > at http level, applied with a single server-level `add_header`. Do **not** move it > into a `location` block: nginx does not inherit `add_header` into a block that > declares its own, so a `location ~* \.(html|css|js)$` setting only `Cache-Control` > silently drops the CSP / HSTS / X-Frame-Options / nosniff headers for exactly those > files. After deploying, confirm both are present on one response: > > ```bash > curl -sI https://wp-suite.company.local/work-package-suite.html > | grep -Ei 'cache-control|content-security-policy' > ```