d78bf52182bde0eaa253b441507129a77bc46cde
4
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 saqib mirandClaude Sonnet 5
|
82d02bcdcd |
feat: implement AI support agent (005) — diagnosis, tools, runbooks, verification
Real Anthropic Claude integration per explicit product decision: a ticket's AI session diagnoses the problem via a structured-output call, applies a DB-configurable confidence-band policy (FR-005), and on "proceed" reasons and acts through a small permission/risk-gated tool system (FR-011/FR-012), optionally walking a matching runbook step by step with the application — never the model — owning the step index (FR-015/FR-016). Resolution requires real tool evidence, never customer claims alone (FR-018) — verifyProductResolution is a documented fail-closed placeholder mirroring the existing malware-scanner precedent, since no real per-product operational signal exists yet. AISupportSession.status mirrors onto Ticket.status through 003-ticketing's existing AI_ANALYZING/AI_TROUBLESHOOTING/AI_VERIFYING/AI_RESOLVED/ HUMAN_ESCALATION state machine, discovered during planning to have been built anticipating this exact feature. Two circular module dependencies (escalation<->sessions, tools<->sessions) were designed around rather than found as bugs: escalation is a pure summary formatter with no state dependencies of its own, and tools stays a clean leaf module with zero dependency on ai-support/sessions. Ticket creation enqueues the first diagnosis turn via the existing queue infrastructure (off the hot path of the inbound SaaS integration endpoint); a human actor changing ticket status ends the AI session via the event-bus scaffold that existed in this codebase but had never been wired to anything. A real Prisma limitation was found and fixed before it reached tests: compound-unique upsert rejects null for a nullable key column, so AIConfidencePolicy uses find-then-update/create instead, same fix class 004 already used for the same underlying limitation. Adds 9 unit tests (confidence-band, tool-policy-gate, runbook-step- advance) and 6 integration test files, including the two constitution- required standing E2E scenarios. AI-independent tests were run against real Postgres/Redis/MinIO (88 passed, 0 failed across the full suite, including every pre-existing 002/003/004 test). The AI-dependent tests compile and skip cleanly via describe.skipIf but were not run against a live model — no ANTHROPIC_API_KEY was available in this session; a real key must be supplied before this feature can actually run. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> |
||
|
saqib mirandClaude Sonnet 5
|
2edfbacf82 |
feat: implement ticket creation, messages & attachments (003-ticketing)
Implements all 39 tasks from specs/003-ticketing/tasks.md across all
three user stories -- Phase 5 of the roadmap.
Schema (prisma/schema.prisma + migration):
- Ticket (code, status, version for optimistic concurrency,
idempotencyKey, customerId FK), Problem, TicketMessage,
TicketAttachment per docs/06, with Product/Category/
CustomerReference back-relations.
User Story 1 -- ticket/problem creation (P1, MVP):
- Explicit 12-state lifecycle adjacency table
(ticket-state-machine.ts), not "any transition allowed."
- Ticket code generation (<PRODUCT_CODE>-<YEAR>-<SEQUENCE>) scoped
by the actual code prefix, not productId -- see the collision bug
fixed below.
- Idempotency-key enforcement via atomic create-then-catch-conflict
(never a read-then-write race), completing the FR-012 placeholder
from 002-saas-integration.
- Explicit-reference-only recurring-problem linking (no fuzzy
matching -- that's a future AI-support concern).
- POST /v1/support/requests (002-saas-integration) now creates a
real ticket instead of echoing context back.
- PATCH /tickets/:id/status with expectedVersion-based optimistic
concurrency (409 on stale version, 400 on an invalid transition).
User Story 2 -- typed messages (P2):
- Message type -> visibleToCustomer mapping is a fixed constant map,
never caller-supplied; customer-scoped reads filter at the query
layer so an internal note is never fetched, not just hidden.
- POST/GET /tickets/:id/messages (customer-scoped) and
GET /agent/tickets/:id/messages (agent-scoped).
User Story 3 -- attachment pipeline (P3):
- Presigned-PUT upload (new getPresignedUploadUrl on the existing
storageService) -- file bytes never transit this API.
- A MalwareScanner interface with a fail-closed placeholder
(UnimplementedPlaceholderScanner) since no scanner exists in this
stack -- it always reports 'infected', never silently 'clean'.
- The existing attachments-queue job stub now actually calls the
scanner and updates scanStatus; registerAttachmentWorker() is
wired into bootstrapQueue() (previously defined but never called).
- Downloads are gated on scanStatus === 'clean' -- currently always
refused until a real scanner replaces the placeholder.
- MinIO added to docker-compose.{test,development}.yml for local/CI
S3-compatible storage, matching doc 04's explicit guidance.
Two real bugs found and fixed via integration testing against a
live Postgres/Redis/MinIO (not just typechecked):
- Ticket codes could collide across different products: the
sequence counter was scoped by internal productId, but the code
column's uniqueness is global, and deriveProductCode's 4-character
truncation means different products can share a prefix. Fixed by
counting against the actual code prefix instead.
- Three existing 002-saas-integration integration tests' cleanup
started failing an FK RESTRICT check once ticket creation was
wired in (deleting a Product before the Ticket/Problem that now
reference it). Fixed their afterAll ordering.
All 9 integration test files (24 tests, spanning this feature and
the pre-existing suite) verified passing against real Postgres,
Redis, and MinIO, including a genuine presigned-PUT/GET round trip.
Full quality gate (typecheck/lint/format/architecture/unit tests)
passes.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
||
|
saqib mirandClaude Sonnet 5
|
8d5731340d |
feat: implement SaaS product integration trust boundary (US1 MVP)
Implements tasks T001-T020 from specs/002-saas-integration/tasks.md (Setup, Foundational, and User Story 1 - the P1 MVP: every inbound request is authenticated and trusted before anything happens). User Story 2 (admin onboarding/rotation/revocation) and User Story 3 (rate limiting) are not yet implemented (T021-T032 remain). Schema (prisma/schema.prisma + initial migration): - Replace the placeholder Product model (leftover starter-template scaffolding: code/description/ProductStatus enum) with the real docs/06-database-schema.md shape (externalProductId, supportEnabled, status). - Add ProductIntegration (credential ref, rotation/revocation state, allowed scope, per-integration/per-user rate limits) and CustomerReference models. - Align AuditLog to docs/06's shape (actor/actorType/entityType/ entityId/reason/metadata) -- the placeholder shape had no fields to satisfy this feature's audit requirements. Auth: - HMAC-signed short-lived tokens (issue/verify) with jti-based replay defense via Redis and a bounded clock-skew tolerance. - Credential secrets are AES-256-GCM encrypted at rest (new required INTEGRATION_CREDENTIAL_ENCRYPTION_KEY env var) since no secret manager exists in this stack yet -- see research.md "Credential storage". - New product-integration-auth.plugin.ts Fastify plugin runs the validation order in contracts/inbound-request-contract.md and populates request.reqContext only on full success; every attempt (success or failure) is audit-logged without ever persisting the raw token/credential. Unregistered product and invalid credential return an identical response (FR-010). - New POST /v1/support/requests endpoint exercises the boundary end-to-end (ticket creation itself is a future feature). Also: - Fix docker-compose.test.yml's container_name collisions -- discovered while testing this change concurrently is now covered by an app-level regression test (separate commit). - Fix test:unit to scope to tests/unit only (it was running the entire tests/** glob including integration tests) -- this feature's new integration test makes real Prisma/Redis calls, unlike the prior instantiation-only checks, so the existing glob-scoping gap became actually harmful. - Update Jenkinsfile with the new required credential. Verified: full quality gate (typecheck/lint/format/architecture/ unit tests) passes; all of User Story 1's quickstart scenarios manually verified end-to-end against a live server + Postgres + Redis; the new integration test suite verified against a live database (not run as part of `npm test`, matches existing test:integration convention). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> |
||
|
saqib mirandClaude Sonnet 5
|
2093898198 |
fix: stop tracking .env files with committed secrets
.env.development had a real Postgres/Redis password and JWT secret committed in plain text; .env.test and .env.prod were tracked too. Untrack all .env* files going forward and add .env.example as the onboarding template instead. Note: the exposed dev credentials are still in git history and must be rotated separately. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> |