Files
slp-modular-cms/aidlc-docs/features/contact-forms-module/construction/phase-1-handoff.md
T

21 KiB

Phase 1 handoff (CMS side) — for the frontend work

Branch feature/contact-forms-module (worktree ..\SlpModularCms-contact-forms), base master d9cc539, phase 1 units CMS-U0, U1, U1b, U2, U3, U3b, U4 all committed (last commit 44f161a). Nothing pushed, merged or deployed. Phase 2 (CMS-U5 mail + retry, CMS-U6 "Berichten" UI, CMS-U7 PUT /availability/status) waits for the merge of feature/identity-email-notifications and a rebase.

The machine-readable source for types is openapi/slpsoftware.v1.json (one file per host, named after the host; regenerate with UPDATE_OPENAPI=1 dotnet test --filter OpenApiSnapshot). Per-unit details: construction/cms-u*/code/summary.md.

All paths are under /api/v1. "Gate" = the availability gate: routes without the public marker answer 503 to visitors while the CMS is off or in maintenance; a verified Owner/Administrator always gets through. 401 = no/invalid login, 403 = logged in without Administrator/Owner. JSON is camelCase.

1. Routes

Offerings (module Offerings)

Method + path Auth Success Other status codes
GET /offerings public (marker, survives a closed gate); limiter offerings-public 120/60 s 200 array of offerings 429
POST /offerings AdminOnly 201 + Location: /api/v1/offerings/{id} + offering 400 (framework validation shape; bad categoryId = errors.categoryId=["invalid"]), 409 slug taken, 401, 403, 503
PUT /offerings/{id:guid} AdminOnly 200 offering (full replace; omitting categoryId clears the category) 400, 404, 409, 401, 403, 503
DELETE /offerings/{id:guid} AdminOnly 204 (soft delete) 404, 401, 403, 503
PUT /offerings/order body { "orderedIds": [guid,...] } AdminOnly 204 400 problem unless the list holds every existing offering exactly once, 401, 403, 503

Offering JSON: { id, slug, title, description, price, priceNote, features[], ctaLabel, featured, displayOrder, categoryId|null, category|null }. category = the category name (a rename shows up at once; no response caching in the code); categoryId and displayOrder are additive fields the SLP site can ignore. Create/update body: title, description, price, priceNote, features[], ctaLabel, featured, slug?, categoryId? (categoryId must exist; null clears).

Offering categories (CMS data, nothing seeded)

Method + path Auth Success Other
GET /offerings/categories AdminOnly 200 [{ id, name, sortOrder, offeringCount }] 401, 403, 503
POST /offerings/categories { name } (1-100) AdminOnly 201 + Location + category 400, 409 name taken (case-insensitive)
PUT /offerings/categories/{id:guid} { name } AdminOnly 200 category 400, 404, 409
DELETE /offerings/categories/{id:guid} AdminOnly 204 404, 409 while live offerings use it

Contact — public (module Contact)

Both answer 404 with an empty body while the form is closed (host switch OR CMS switch off); the website then shows its mailto fallback. Both stay reachable when the CMS is disabled.

Method + path Auth Success Other
GET /contact-messages/token public (marker); limiter contact-token 20/600 s 200 { token, minSeconds, captcha: null | { provider, siteKey } } 404 closed, 429
POST /contact-messages headers Idempotency-Key: <uuid> (required), body { name, email, subject, message, hpCheck, token, captchaToken? } public (marker); limiter contact-form 5/600 s 202 { "status": "received" } — identical for genuine, spam and replay 400 problem+json (errors per field), 404 closed, 413 problem (> 16384 bytes), 429

Limits: name 1-100, email 1-254 (regex ^[^\s@]+@[^\s@]+\.[^\s@]{2,}$), subject 1-100 (opaque value chosen by the site), message 1-2000; all trimmed; control characters rejected (line breaks and tabs allowed in message). hpCheck is the honeypot, must be empty. 400 shape: { type, title, status, errors: { <field>: [code] } }, codes required | tooLong | invalid | expired; fields name, email, subject, message, token, idempotencyKey, body. A forged, other-site or expired token is always errors.token=["expired"] (fetch a fresh token and retry once). Replays with the same key within 24 h get the same 202 without a second message; a 400 does not use up the key.

Contact — management (AdminOnly; not public; behind the gate)

All answer 404 with an empty body while the host switch Contact:Enabled is off (the tables do not exist then). The CMS switch never closes them.

Method + path Success Other
GET /contact-messages?status=&page=&pageSize= 200 { items[{id,receivedAt,name,email,subject,status,mailStatus}], page, pageSize, totalCount, unreadCount }; default list = new,read,archived (no spam), status=spam shows spam apart; pageSize 1-100 (default 25); unreadCount = new only, never spam 400 (errors.page/pageSize/status), 401, 403, 404, 503
GET /contact-messages/{id:guid} 200 detail (list fields + message, mailAttempts, mailError, retentionUntil) 404
PATCH /contact-messages/{id:guid} { "status": "new"|"read"|"archived" } 200 detail (on a spam message = "not spam": normal retention from now) 400 errors.status=["invalid"], 404
DELETE /contact-messages/{id:guid} 204 (hard delete) 404
POST /contact-messages/{id:guid}/resend (limiter contact-resend 5/600 s) 200 { mailStatus, mailAttempts, mailed } (mailed:false until CMS-U5) 404, 409 spam message, 429
GET /contact-settings 200 { enabled, hostEnabled } (enabled = CMS switch, default false) 401, 403, 404, 503
PUT /contact-settings { "enabled": true|false } 200 same shape 400 errors.enabled=["invalid"] for anything that is not a JSON boolean, 401, 403, 404, 503

status: new|read|archived|spam|converted; mailStatus: pending|sent|failed|notApplicable.

Unchanged but relevant

PUT /availability/status (AdminOnly; body { newStatus, reason }; 200 with the updated resource in the GET shape, 409 master-controlled) replaced POST /availability/admin/status (CMS-U7, no alias; API and admin app must be released together). GET /availability/status stays public; the gate exempts it for GET/HEAD only.

Key Default Notes
Contact:Enabled false host switch; release condition, see § 3
Contact:SiteId slpsoftware (code default default) stored on rows, in the token purpose; never from a request
Contact:MinSeconds 3 returned as minSeconds; faster = spam
Contact:TokenLifetimeMinutes 120
Contact:RetentionMonths 12 fixed per row at insert
Contact:SpamRetentionDays 30
Contact:MaxSpamRowsPerDay 200 per site per UTC day; beyond it spam is answered but not stored
Contact:MaxBodyBytes 16384 413 beyond
Contact:ProblemTypeBase none -> about:blank {base}/validation when set
Contact:Captcha:Enabled false
Contact:Captcha:Provider, :SiteKey none returned in the token response when enabled
Contact:Captcha:Secret none environment / secrets manager only, never committed
RateLimiting:ContactForm 5 / 600 PermitLimit / WindowSeconds
RateLimiting:ContactToken 20 / 600
RateLimiting:ContactResend 5 / 600
(removed) Offerings:Categories no longer exists; categories are data

CMS switch (not configuration): table ContactSettings, PUT /contact-settings, default off. Mail recipient/sender settings arrive with CMS-U5 (environment only, no addresses in committed config).

3. Release / rebase checklist (do in this order, after Bryan merges feature/identity-email-notifications)

  1. Rebase feature/contact-forms-module onto the merged base. Expected conflicts: AvailabilityMiddleware.cs (one block), AddCmsRateLimiting (OnRejected), appsettings files, host composition, maybe EF snapshots (own contexts, so none expected; regenerate our migration only if a snapshot conflicts). Re-run all test projects.
  2. Per-client rate limiting: in Modules.Contact/Services/ContactSeams.cs, class ContactRateLimitPartition, change public static string KeyFor(HttpContext context) => GlobalKey; to => ClientPartitionKey.For(context); (add the ASF using; drop GlobalKey if unused). In Contact.Tests/ContactRetentionAndPartitionTests.cs, test OnTheBaseBranch_EveryCallerSharesOneBucket becomes "two different client addresses give two different keys". Until this is done every visitor shares one bucket per policy (5 form submissions per 10 minutes for the whole site), which is why the module is off by default.
  3. ForwardedHeaders:KnownProxies configured on the host (environment).
  4. Phase 2 units (CMS-U5 mail + retry job, CMS-U6 UI, CMS-U7) on the rebased branch; give the Andrea host the Contact module reference and its own openapi/andreaslopfotografie.v1.json when it is in scope.
  5. Only then set Contact:Enabled=true on the host (environment or its appsettings).
  6. Switch the CMS switch on: it is OFF by default, so Bryan must open the form himself (screen "Berichten" in CMS-U6, or PUT /api/v1/contact-settings {"enabled":true}). Until both switches are on the website shows its mailto fallback.
  7. Nothing is deployed to production without Bryan; no push/merge/PR happened.

4. Known deviations from contract v0.3.x and design decisions to adopt

  • Admin routes are RESTful on the resource (/contact-messages, /offerings, /contact-settings), no admin segment, per Bryan's decision; old offerings/admin... routes and move-up/move-down are gone (reorder via PUT /offerings/order).
  • One offering DTO for everybody (OfferingAdminDto removed): adds displayOrder, categoryId; category is the name.
  • Categories are CMS data (table + CRUD), not Offerings:Categories.
  • Honeypot JSON name is fixed hpCheck (no configurable name); a missing hpCheck counts as empty.
  • Unbindable body (malformed JSON, wrong types) -> 400 with errors.body=["invalid"] in the same problem shape (the contact controllers are deliberately not [ApiController]).
  • Contact:ProblemTypeBase configurable, default about:blank (instead of a fixed URL).
  • Two-layer switch: host Contact:Enabled and the CMS switch; both default off; 404 empty body when closed.
  • GET /contact-settings returns the extra read-only hostEnabled.
  • Un-spam (PATCH spam -> new/read/archived) extends retention to the normal period; resend of spam = 409.
  • Management routes answer 404 while the host switch is off; an Owner/Administrator passes the availability gate while the CMS is disabled (existing design), visitors get 503.
  • Mail: stored first; the mail seam (IContactMailSender) is a no-op until CMS-U5, so messages stay pending and resend reports mailed:false.
  • Rate limits are global buckets on this base until the checklist step 2 (documented, enforced by the default-off switches).
  • Privacy: no IP address is stored; Sentry error events carry the caller's IP (SendDefaultPii) but never form content — mention error reporting in the privacy statement. Backups may keep deleted messages longer than the retention period — mention or shorten backup retention.

5. Tests and tooling worth knowing

  • Pipeline tests (Contact.Tests/Pipeline, Api.Tests) need ConnectionStrings__DefaultConnection pointing at a MariaDB/MySQL server on which databases may be created (CI provides it: .gitea/workflows/continuous_integration.yaml). Each fixture creates and drops its own database. Unit tests only: dotnet test --filter "FullyQualifiedName!~Pipeline".
  • Frontend (frontend/): pnpm lint, npx tsc -b, pnpm test (278 tests at CMS-U1b). 15 pre-existing files under features/offerings fail prettier --check (tech debt, not touched).
  • The admin frontend work of phase 1 is only the offerings part (new routes, useMoveOffering via PUT /offerings/order, categories dialog + select, query key ['offerings','list']).
  • Last test run (CMS-U4): Contact.Tests 223, Core 205, Availability 95, Offerings 86, Identity 37, Master 57, Api.Tests 4 — all green.

6. Rebase result (2026-11-10)

Rebased onto origin/master e671078 (merge of PR #11: identity-email-notifications + andreaslop-site-hosting). Conflicts: aidlc-docs/active-features.md (kept master's table, appended our row), AvailabilityMiddleware (kept their CheckAdmin / expired-token 401 behaviour, our marker block before it), AddCmsRateLimiting.OnRejected (kept their Retry-After logic for sliding windows, added the 60 s fallback for policies without a wait and the problem+json body), SlpModularCms.sln (took master's, re-added the two Contact projects). No EF snapshot conflict (own contexts). Checklist steps 1 and 2 are DONE: ContactRateLimitPartition.KeyFor now returns ClientPartitionKey.For(context) and its test asserts different buckets per address. Still open: step 3 (ForwardedHeaders:KnownProxies per host), step 5 (Contact:Enabled=true), step 6 (CMS switch). GET /System/capabilities now carries [PublicSiteEndpoint] (never 503; separate commit). The pipeline tests for the management routes log in with a real user (the host validates every access token against its user and security stamp).

7. CMS-U5 additions (contact mail)

  • Recipient address: since CMS-U11 the site setting wins over the env var Contact__RecipientEmail (see section 9); the env var stays the default/fallback (SLP: environment only, like ASF). Optional Contact__Mail__SweepSeconds|MaxAttempts|ClaimMinutes|BatchSize, Contact__Mail__BackoffMinutes__0..n.
  • Mail itself needs the host's normal Email:* configuration; without it messages stay pending with mailError: "disabled".
  • Detail: nextMailAttemptAt; resend: mailError. New migration ContactMailClaim.

8. CMS-U9 additions (Friendly Captcha v2 and the ASF host)

Captcha contract (no wire change)

  • GET /contact-messages/token -> captcha: null while off, otherwise { "provider": "friendlycaptcha", "siteKey": "<public site key>", "apiEndpoint": "global" | "eu" }. apiEndpoint (additive, default global) is the widget's data region and follows Contact:Captcha:Region, so switching region needs no website rebuild. Nothing else about the captcha (API key, verify URL) is ever returned.
  • POST /contact-messages: captchaToken = the widget's response string (the widget also puts it in the form field frc-captcha-response). While the captcha is on, a missing, empty or rejected value is treated as spam: answer 202 {"status":"received"} identical to a genuine message, stored with status spam (never a 400, so a bot learns nothing). While the captcha is off the field is ignored.
  • Verify (server side, Friendly Captcha v2): POST https://global.frcapi.com/api/v2/captcha/siteverify (EU-only: https://eu.frcapi.com/api/v2/captcha/siteverify), header X-API-Key: <API key>, JSON { "response": "...", "sitekey": "..." }; success:true passes, success:false (HTTP 200 response_invalid|response_timeout|response_duplicate, or 400 response_missing|bad_request) = spam. Checked against developer.friendlycaptcha.com on 2026-11-10.
  • Fail-open: timeout (default 3 s), network error and 5xx let the visitor through (warning captcha_unavailable, kind only). Our own wrong API key or site key (HTTP 401, or 400 with error_code: sitekey_invalid) also lets the visitor through but logs an error captcha_misconfigured. The response token, the API key and the site key are never logged.

Configuration (per site)

Key Notes
Contact:Captcha:Enabled default false
Contact:Captcha:Provider friendlycaptcha (only supported value; case-insensitive)
Contact:Captcha:SiteKey public; sent to the browser
Contact__Captcha__Secret the API key; environment variable / secrets manager only, never committed
Contact:Captcha:Region global (default) or eu; decides the verify host and the apiEndpoint announced to the widget. Unknown value = host refuses to start
Contact:Captcha:VerifyUrl optional override, normally empty (derived from Region); if set it must be https and match the Region's host (global.frcapi.com / eu.frcapi.com), otherwise the host refuses to start
Contact:Captcha:TimeoutSeconds default 3 (1-30)

With Contact:Enabled=true and Captcha:Enabled=true the host refuses to start when provider, site key or secret is missing/invalid (the message names the setting only, never a value).

Widget (website session) and CSP

  • Widget: Friendly Captcha v2 SDK with sitekey = siteKey from the token response; for the EU endpoint the widget needs apiEndpoint: "eu" (data-api-endpoint="eu"). The response goes into captchaToken. Exact SDK package/events: to be verified by the website session against the SDK docs.
  • CSP (decision: exact origins, no wildcard): the v2 widget renders in iframes and the provider's CSP guide requires frame-src for *.frcapi.com. New option SecurityHeaders:AllowedSiteFrameOrigins (StrictSite only; empty = header byte-identical to before; origins only, validated at start-up; the admin and relaxed policies never get it). ASF values (owner decision: main host plus two reserve hosts, exact, no wildcard): SecurityHeaders__AllowedSiteFrameOrigins__0=https://global.frcapi.com, __1=https://global0.frcapi.com, __2=https://global1.frcapi.com. On the switch to the EU region use https://eu.frcapi.com, https://eu0.frcapi.com, https://eu1.frcapi.com instead (together with Contact__Captcha__Region=eu). The resulting directive is frame-src 'self' https://global.frcapi.com https://global0.frcapi.com https://global1.frcapi.com. No connect-src/script-src additions are documented by the provider; a browser test with the real widget is still needed.
  • Privacy (provider statements, v2 docs): "Friendly Captcha does not rely on user tracking or cookies for its protection service". Region global: the visitor's IP address can be processed outside the EU. Region eu (dedicated endpoint "hosted in Germany"; paid Advanced/Enterprise plan): "personal information (i.e. visitor IP addresses) never leave the EU". The site's privacy page must name the captcha provider, the processing of the visitor's IP address for the challenge, and which region is in use; commercial use needs a paid plan with the provider. For a test or a launch without a paid plan keep Contact:Captcha:Enabled=false.

ASF host (Api.AndreaSlopFotografie)

  • References Modules.Contact; section Contact in its appsettings.json (Enabled:false, SiteId: andreaslop), the three contact rate-limit policies. Capabilities now list Contact; the module does nothing and touches no database while Contact:Enabled=false.
  • Release conditions (also in the comment of that section): ForwardedHeaders__KnownProxies__0 = reverse proxy (IIS) address first (otherwise the whole site shares one bucket); Contact__RecipientEmail (env only); mail via Email__Smtp__*, App__PublicBaseUrl; sender = Email:FromAddress or the address chosen in the site settings from Email:AllowedFromAddresses (set by environment for ASF); captcha settings above; then Contact__Enabled=true and the admin switch (Berichten).
  • Host tests (Api.AndreaSlopFotografie.Tests/ContactHostTests.cs, real MariaDB): closed by default; public routes stay reachable (200/202) while the availability gate is "not available" and the management routes answer 503; per-client buckets with a known proxy vs one shared bucket without.

9. CMS-U11 additions (recipient address from the site settings)

Order of precedence for the address contact mails are sent to: site setting (CMS admin, Berichten) > Contact__RecipientEmail (environment) > none. The setting changes without a restart (read on every send); an invalid stored value is ignored (falls back). Table ContactSettings gets a nullable RecipientEmail (migration ContactSettingsRecipient, additive).

Method + path Auth Notes
GET /contact-settings AdminOnly now { enabled, hostEnabled, recipientEmail: string|null, recipientSource: "setting"|"config"|"none" }. recipientEmail is only the CMS setting; the address from the environment is never returned (source config = "set by the server configuration").
PUT /contact-settings/recipient { "recipientEmail": string|null } AdminOnly 200 same shape as GET; null/blank clears the setting (back to the environment default); trimmed; EmailAddressRules; 400 problem+json errors.recipientEmail = ["invalid"] or ["tooLong"] (> 254), never echoing the value; 401/403/404/503 as the other management routes. Not part of any public route or OpenAPI example.
PUT /contact-settings { enabled } AdminOnly unchanged.

Without any address (recipientSource: "none") the host still starts and the form can still be switched on (warn, not block): messages are stored, the mail stays pending with mailError: "recipient_not_configured" (not counted as an attempt) and goes out by itself within seconds once an address is set (setting the address wakes the mail sender). The admin shows a warning banner on the switch card. Release condition for SLP/ASF: the address no longer has to be in the environment when it is set in the admin.