docs: update project documentation to reflect security remediation

Update README, architecture guide, and feature inventory to document all
changes from the security remediation pass (17 tickets):

- README: Add CI pipeline section (race detection, golangci-lint,
  govulncheck, per-layer coverage thresholds), CORS deny-by-default
  behavior, input validation, SSRF protection, scheduler concurrency
  safety. Update test count to 1050+. Add race detection and govulncheck
  to development commands.

- Architecture guide: Update testing strategy with scheduler tests, fuzz
  tests, and revised CI pipeline description. Add security model sections
  for input validation, CORS, and concurrency safety. Update test count.

- Feature inventory: Document CORS deny-by-default behavior.

- SECURITY_REMEDIATION.md: New file documenting all 17 remediated tickets
  with CWE classifications, before/after behavior, 3 deferred tickets
  with rationale, CI pipeline changes, and breaking CORS change.

Missing docs flagged as future additions:
- Formal threat model document
- Disaster recovery runbook
- Version upgrade guide
- Capacity planning benchmarks

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

docs/architecture.md

@@ -713,6 +713,18 @@ In addition to application-level audit events, certctl records every HTTP API ca
 
 Audit recording is async (via goroutine) so it never blocks the HTTP response. If audit persistence fails, the error is logged immediately — the API call still succeeds. The middleware sits after the auth middleware in the stack so the actor identity is available from context.
 
+### Input Validation and SSRF Protection
+
+All shell-facing inputs (connector scripts, domain names, ACME tokens) are validated through `internal/validation/command.go` before reaching shell execution. `ValidateShellCommand()` denies all shell metacharacters. `ValidateDomainName()` enforces RFC 1123. `ValidateACMEToken()` restricts to base64url characters. The network scanner filters reserved IP ranges (loopback, link-local including cloud metadata 169.254.169.254, multicast, broadcast) to prevent SSRF, while preserving RFC 1918 private ranges for legitimate internal scanning.
+
+### CORS
+
+CORS uses a **deny-by-default** posture: when `CERTCTL_CORS_ORIGINS` is empty, no CORS headers are set and only same-origin requests can read responses. Operators must explicitly configure allowed origins. This prevents accidental exposure of the API to cross-origin requests in production.
+
+### Concurrency Safety
+
+The background scheduler uses `sync/atomic.Bool` idempotency guards on all 6 loops — if a tick fires while the previous iteration is still running, it skips. A `sync.WaitGroup` tracks all in-flight goroutines. `WaitForCompletion(timeout)` blocks during shutdown until all work finishes or the timeout expires, preventing state corruption from mid-flight database operations during process exit.
+
 ### Logging
 
 All logging throughout the service layer uses Go's `log/slog` package for structured, queryable logs. This replaces ad-hoc `fmt.Printf` statements with consistent key-value logging that includes request context, operation names, and error details. Agents also implement exponential backoff on network failures to gracefully handle temporary connectivity issues with the control plane.
@@ -895,7 +907,7 @@ This data flow is pull-based and non-blocking. Agents discover at their own pace
 
 ## Testing Strategy
 
-certctl uses a layered testing approach aligned with the handler → service → repository architecture, with 900+ tests across five layers (service, handler, integration, connector, and frontend). The goal is high-confidence regression prevention at the service and handler layers, where the most complex business logic lives, combined with integration tests that exercise the full request path from HTTP to database.
+certctl uses a layered testing approach aligned with the handler → service → repository architecture, with 1050+ tests across six layers (service, handler, integration, connector, frontend, and scheduler). The goal is high-confidence regression prevention at the service and handler layers, where the most complex business logic lives, combined with integration tests that exercise the full request path from HTTP to database.
 
 **Service layer unit tests** (`internal/service/*_test.go`) — ~238 test functions across 15 files with mock repositories. These test all business logic in isolation: certificate CRUD with validation, certificate revocation (success, already-revoked, archived, invalid reason, all RFC 5280 reason codes, issuer notification, notification service integration, OCSP/CRL generation), agent lifecycle (registration, heartbeat, CSR submission with both keygen modes), job state machine (creation, processing, cancellation, retry logic), policy evaluation (all 5 rule types, violation creation), renewal and issuance flow (server-side and agent-side keygen paths), notification deduplication (threshold tag matching, channel routing), team/owner/agent group CRUD with pagination and audit recording, issuer service CRUD with connection testing, and the issuer connector adapter (type translation between connector and service layers including revocation). Mock repositories are simple structs with function fields, avoiding heavy mocking frameworks — this keeps tests readable and avoids coupling to mock library APIs.
 
@@ -907,11 +919,15 @@ certctl uses a layered testing approach aligned with the handler → service →
 
 **CLI tests** (`internal/cli/client_test.go`) — 14 tests covering all 10 CLI subcommands with httptest mock servers, PEM parsing for bulk import, auth header verification, and JSON/table output formatting.
 
-**CI pipeline** (`.github/workflows/ci.yml`) — Two parallel jobs: Go (build, vet, test with coverage, coverage threshold enforcement) and Frontend (TypeScript type check, Vitest test suite, Vite production build). The Go job runs all tests with `-coverprofile`, then enforces coverage thresholds: service layer must be at least 30% (current: ~35%) and handler layer must be at least 50% (current: ~63%). These thresholds act as regression floors — they can only go up. The service layer threshold is deliberately lower because much of the service code depends on postgres repositories and external connectors that require real infrastructure to test meaningfully. Connector tests are included via `./internal/connector/issuer/...` and `./internal/connector/target/...` (covers Local CA, ACME, step-ca, NGINX, Apache, HAProxy, Traefik, and Caddy packages with unit tests for certificate signing logic, DNS solver, issuer validation, and deployment flows). The Frontend job runs `npx vitest run` between the TypeScript check and production build steps.
+**CI pipeline** (`.github/workflows/ci.yml`) — Two parallel jobs: Go (build, vet, race detection, static analysis, vulnerability scanning, test with coverage, coverage threshold enforcement) and Frontend (TypeScript type check, Vitest test suite, Vite production build). The Go job runs `go test -race` on service, handler, middleware, and scheduler packages to catch data races. It runs `golangci-lint` with 11 linters (errcheck, govet, staticcheck, unused, gosimple, ineffassign, typecheck, gocritic, gosec, bodyclose, noctx) configured in `.golangci.yml`. It runs `govulncheck ./...` to scan dependencies for known CVEs. Coverage thresholds are enforced per-layer: service 60%, handler 60%, domain 40%, middleware 50%. These thresholds act as regression floors — they can only go up. Connector tests are included via `./internal/connector/issuer/...` and `./internal/connector/target/...` (covers Local CA, ACME, step-ca, NGINX, Apache, HAProxy, Traefik, and Caddy packages with unit tests for certificate signing logic, DNS solver, issuer validation, and deployment flows). The Frontend job runs `npx vitest run` between the TypeScript check and production build steps.
 
 **Connector tests** (`internal/connector/`) — 57 test functions covering issuer, target, and notifier connectors. The Local CA connector has tests for self-signed and sub-CA modes (RSA, ECDSA, config validation, non-CA cert rejection). The ACME DNS solver has 10 tests for script-based DNS-01 and DNS-PERSIST-01 challenges (6 DNS-01 tests + 4 DNS-PERSIST-01 tests covering `PresentPersist` success, no-script error, script failure, and wildcard domain handling). The step-ca connector has tests with a mock HTTP server for issuance, renewal, revocation, and error paths. The OpenSSL/Custom CA connector has 14 tests covering config validation, issuance success/failure/timeout, renewal, revocation, and CRL generation. The NGINX target connector has 13 tests covering config validation, certificate deployment (file writing, permissions, validate/reload commands), and deployment validation. Apache httpd and HAProxy connectors each have 3 tests covering config validation, deployment, and validation flows. Traefik and Caddy connectors have tests covering file-based deployment and (for Caddy) dual-mode API/file configuration. Notifier connector tests span 20 tests across Slack (5), Teams (4), PagerDuty (6), and OpsGenie (5) — verifying channel identity, payload formatting, HTTP error handling, connection failures, auth headers, and configuration defaults.
 
-**What's not tested and why:** Postgres repository implementations (`internal/repository/postgres/`) require a real database and are tested only through integration tests, not unit tests. Target connectors for F5 BIG-IP and IIS are interface stubs (implementation planned for a future release). Scheduler loops are time-dependent and tested manually during development. The ACME connector requires a real ACME server (tested manually against Let's Encrypt staging). These are all candidates for future expansion as the test infrastructure matures.
+**Scheduler tests** (`internal/scheduler/scheduler_test.go`) — Tests for idempotency guards (`sync/atomic.Bool` CompareAndSwap prevents concurrent loop ticks), `WaitForCompletion` success and timeout paths, and multi-loop idempotency.
+
+**Fuzz tests** (`internal/validation/command_fuzz_test.go`, `internal/domain/revocation_fuzz_test.go`) — Go native fuzz tests (`testing/fuzz`) for command validation functions and revocation domain parsing. These exercise `ValidateShellCommand`, `ValidateDomainName`, and `ValidateACMEToken` with random inputs to discover edge cases.
+
+**What's not tested and why:** Postgres repository implementations (`internal/repository/postgres/`) require a real database and are tested only through integration tests, not unit tests — a `testcontainers-go` scaffolding for isolated PostgreSQL instances is planned. Target connectors for F5 BIG-IP and IIS are interface stubs (implementation planned for V3). The ACME connector requires a real ACME server (tested manually against Let's Encrypt staging). These are all candidates for future expansion as the test infrastructure matures.
 
 ## What's Next