mirror of
https://github.com/shankar0123/certctl.git
synced 2026-06-07 16:01:30 +00:00
cbb47aaf5dfda4b093d5ada5f6d01b31c56d808e
13 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 shankar0123
|
cbb47aaf5d |
auth-bundle-1 Phase 11 + 12: RBAC MCP tools + negative-test coverage gate
# Phase 11 — RBAC MCP tools
12 new tools in internal/mcp/tools_auth.go mirroring the Phase-4
+ Phase-7 HTTP surface so operators driving certctl from Claude
/ VS Code / any MCP client get the same management capability
the GUI + CLI already expose:
certctl_auth_me GET /v1/auth/me
certctl_auth_list_roles GET /v1/auth/roles
certctl_auth_get_role GET /v1/auth/roles/{id}
certctl_auth_create_role POST /v1/auth/roles
certctl_auth_update_role PUT /v1/auth/roles/{id}
certctl_auth_delete_role DELETE /v1/auth/roles/{id}
certctl_auth_list_permissions GET /v1/auth/permissions
certctl_auth_add_permission_to_role POST /v1/auth/roles/{id}/permissions
certctl_auth_remove_permission_from_role DELETE /v1/auth/roles/{id}/permissions/{perm}
certctl_auth_list_keys GET /v1/auth/keys
certctl_auth_assign_role_to_key POST /v1/auth/keys/{id}/roles
certctl_auth_revoke_role_from_key DELETE /v1/auth/keys/{id}/roles/{role_id}
Each tool routes through the existing HTTP client (no parallel
business logic), so permission gates fire server-side: a
non-admin caller's MCP tool invocation returns whatever 403 the
underlying HTTP handler emits, fenced via errorResult for LLM-
prompt-injection defense.
Input types in internal/mcp/types.go (AuthRoleIDInput,
AuthCreateRoleInput, AuthUpdateRoleInput,
AuthRolePermissionGrantInput, AuthRolePermissionRevokeInput,
AuthAssignKeyRoleInput, AuthRevokeKeyRoleInput) carry
jsonschema descriptions so the MCP consumer's tool catalogue
shows operator-friendly hints.
internal/mcp/tools_auth_test.go ships 14 tests:
- TestAuthMCP_AllToolsRegister (registration must not panic)
- TestAuthMCP_PathsAndMethods (table-driven, 12 rows pinning
each tool's HTTP method + URL)
- TestAuthMCP_ForbiddenSurfacesFencedError (12 tools × 403
mock → error surface)
internal/mcp/tools_per_tool_test.go's allHappyPathCases extended
with the 12 new rows so the in-memory dispatch coverage gate
(TestMCP_RegisterTools_DispatchableToolCount) stays green at the
new total of 139 registered tools.
Re-derived total via 'grep -cE "gomcp\.AddTool\(" internal/mcp/tools*.go':
133 (121 in tools.go + 12 in tools_auth.go).
# Phase 12 — negative-test coverage gate
Audit of the prompt's 12 negative-test paths against existing
coverage:
1. Missing actor → 401 ✓ TestRequirePermission_NoActorReturns401, TestRBACGate_NoActorReturns401
2. No roles → 403 ✓ TestRequirePermission_DeniedActorReturns403, TestRBACGate_AuditorRole_403sOnAdminRoutes
3. Role lacks specific perm → 403 ✓ same suite
4. Wrong scope → 403 ✓ TestAuthorizer_SpecificScopeMatchesExactID (wrongID arm)
5. Self-grant w/o auth.role.assign → 403 ✓ TestActorRoleService_GrantRequiresAuthRoleAssign
6. Bootstrap token wrong → 401 ✓ TestEnvTokenStrategy_WrongTokenReturnsInvalidToken, TestBootstrapHandler_Mint_WrongToken_401
7. Bootstrap used twice → 410 ✓ TestEnvTokenStrategy_OneShotConsumption, TestBootstrapHandler_Mint_TwiceReturns410
8. Bootstrap when admin exists → 410 ✓ TestEnvTokenStrategy_AdminExistsClosesPath, TestBootstrapHandler_Mint_AdminExists410
9. Role delete with assignees → 409 NEW: TestRoleService_DeleteWithActorsAssignedReturns409
10. Profile-edit loophole → gated ✓ TestProfileEdit_RequiresApprovalLoopholeClosed
11. Permission not in catalog → 400 ✓ TestRoleService_AddPermissionRejectsNonCanonical
12. Scope ID for nonexistent resource → 404 (validation deferred — no FK constraint between role_permissions.scope_id and the resource tables; documented for a future bundle)
Filled the gap at #9 with TestRoleService_DeleteWithActorsAssignedReturns409
which pins the repository sentinel pass-through (postgres FK
ON DELETE RESTRICT → repository.ErrAuthRoleInUse → service
returns the sentinel verbatim → handler maps to HTTP 409).
# Coverage gates
.github/coverage-thresholds.yml gains 2 entries:
- internal/auth: floor 85
- internal/service/auth: floor 85
.github/workflows/ci.yml's coverage test command extended with
./internal/auth/... and ./internal/api/router/... so the
threshold check has data to evaluate.
# Protocol-endpoint not-gated test (Category F)
internal/api/router/phase12_protocol_allowlist_test.go (new)
adds 3 router-level invariant tests:
- TestPhase12_ProtocolEndpointsNotGated: AST-walks router.go,
asserts no rbacGate(...) call references a path under any
protocol-endpoint prefix (/acme, /scep, /.well-known/est,
/.well-known/pki/ocsp, /.well-known/pki/crl).
- TestPhase12_IsProtocolEndpoint_CoversCanonicalPrefixes:
pins auth.IsProtocolEndpoint against the canonical prefix
set; if a future protocol lands without lockstep allowlist
update, this fails.
- TestPhase12_RBACGateRoutesAreUnderAPIv1: belt-and-braces —
every rbacGate-wrapped route MUST start with /api/v1/.
Catches accidental cross-prefix wraps.
Complements the existing TestRequirePermission_ProtocolEndpointBypassesGate
(middleware-level) + TestRouter_AuthExemptAllowlist_PinsActualRegistrations
(allowlist drift) so the Category F invariant is pinned at all
three layers (middleware + router + dispatch).
# Verifications
* gofmt clean repo-wide.
* go vet ./... clean.
* staticcheck across internal/auth + handler + router + cli +
service + repository + cmd + domain + mcp: clean.
* go test -short -count=1 green across internal/auth (incl.
bootstrap), internal/api/handler, internal/api/router,
internal/cli, internal/service (incl. auth),
internal/domain/auth, internal/mcp, cmd/server, cmd/cli.
|
||
|
shankar0123
|
99a012e3be |
auth-bundle-1 Phase 0: extract internal/auth/ from middleware package
Bundle 1 / Phase 0: pure refactor splitting auth surface out of internal/api/middleware so Bundle 2 (OIDC + sessions) and the broader RBAC primitive (roles, permissions, scoped grants) have a clean home. Moved to internal/auth/: NamedAPIKey, HashAPIKey, AuthConfig, NewAuthWithNamedKeys, NewAuth, UserKey, AdminKey, GetUser, IsAdmin. Added testfixtures.go (WithActor / WithAdmin / WithActorAdmin) so handler tests don't construct context manually. Stayed in internal/api/middleware/: RequestID, Logging, NewLogging, Recovery, RateLimitConfig, NewRateLimiter (now imports auth.GetUser for per-user keying per audit Category C), CORSConfig, NewCORS, ContentType, CORS, GetRequestID, responseWriter, Chain, audit middleware (now imports auth.GetUser). Updated 22 caller files across cmd/, internal/api/handler/, internal/api/middleware/, internal/mcp/. Existing m008_admin_gate_test.go now scans for auth.IsAdmin( substring; Phase 3 will further evolve to track auth.RequirePermission. Behavior unchanged: all handler / middleware / service / connector / cmd / mcp tests pass with no test-logic edits, only import-path renames. Phase 0 exit criteria: internal/auth/ exists with 6 files; middleware.go went 575 -> 422 lines (auth-related ~150 lines moved out); grep -rE 'middleware\.(GetUser|IsAdmin|UserKey|AdminKey|NamedAPIKey|HashAPIKey|NewAuth)' returns 0 hits; context.WithValue(.*middleware.UserKey/AdminKey) returns 0 hits; go vet ./... clean; go test -short ./... green across all packages tested. Branch: dev/auth-bundle-1. Per cowork/auth-bundle-1-prompt.md, do not merge to master without (1) make verify green, (2) >= 2 external testers confirm, (3) >= 90% coverage on internal/auth/ in .github/coverage-thresholds.yml. |
||
|
shankar0123
|
ff75361553 |
mcp(coverage): add 34 tools across 7 domains to close 2026-05-05 parity audit P1 findings
Closes findings P1-1..P1-35 from the 2026-05-05 CLI/API/MCP↔GUI parity
audit (cowork/cli-gui-parity-audit-2026-05-05/RESULTS.md). Before this
bundle, 35 operator-facing API endpoints had GUI surfaces but no MCP
counterpart — operators using AI assistants for cert lifecycle work in
regulated environments had to drop to curl for approve/reject, health-check
acknowledgement, renewal-policy CRUD, network-scan triggering, discovery
triage, intermediate-CA management, and job verification.
Tool count: 87→121 in tools.go (+34), 6 unchanged in tools_est.go.
Re-derive via grep -cE 'gomcp\\.AddTool\\(' internal/mcp/tools.go
internal/mcp/tools_est.go.
The 7 phases (matching the bundle prompt at
cowork/mcp-coverage-expansion-prompt.md):
Phase A — Approvals (P1-28..P1-31, 4 tools)
list_approvals, get_approval, approve_request, reject_request.
Two-person-integrity contract (ErrApproveBySameActor → HTTP 403)
is preserved automatically: the decided_by actor is derived
server-side from middleware.UserKey, NOT from request body, so
the MCP server's authenticated API-key identity becomes the
audit-trail actor. The MCP input schema deliberately omits any
actor_id field to prevent client-side spoofing.
Phase B — Health Checks (P1-20..P1-27, 8 tools)
list, summary, get, create, update, delete, history, acknowledge.
Mirrors the existing target-resource shape; acknowledge takes
optional 'actor' string captured in the audit row (handler defaults
to 'unknown' if absent).
Phase C — Renewal Policies (P1-1..P1-5, 5 tools)
Standard CRUD against /api/v1/renewal-policies. Distinct from the
legacy 'policy' tools that point at the same path — these expose
the renewal-policy domain explicitly with full alert_channels +
alert_severity_map field shape.
Phase D — Network Scan Targets (P1-14..P1-19, 6 tools)
CRUD + trigger_scan. trigger_network_scan returns the discovery-
scan body so the AI can chain into list_discovered_certificates
filtered by agent_id.
Phase E — Discovery read-side (P1-10..P1-13, 4 tools)
list_discovered_certificates, get_discovered_certificate,
list_discovery_scans, discovery_summary. Complements the
pre-existing claim/dismiss tools (registered alongside Health
historically per the I-2 closure).
Phase F — Intermediate CAs (P1-6..P1-9, 4 tools)
list, create (root + child via discriminator on body shape), get,
retire. The handler is admin-gated via middleware.IsAdmin; the
least-privilege boundary is enforced at the API layer (HTTP 403
for non-admin Bearer callers) — not by transport carve-out.
Phase G — Verification + deployments (P1-32, P1-34, P1-35, 3 tools)
list_certificate_deployments, verify_job, get_job_verification.
P1-33 (POST /api/v1/agents/{id}/discoveries) is intentionally
excluded — machine-to-machine push channel for agents reporting
filesystem-scan results, not an operator-driven flow. Documented
inline in the RegisterTools dispatch.
Implementation:
- 14 new input types in internal/mcp/types.go with jsonschema struct
tags driving LLM tool discovery.
- 7 register* functions in internal/mcp/tools.go each handling one
phase, wired into RegisterTools dispatch in declaration order.
- 34 new entries in tools_per_tool_test.go::allHappyPathCases —
the existing in-process MCP harness (TestMCP_AllTools_HappyPath +
TestMCP_AllTools_ErrorPath + TestMCP_RegisterTools_DispatchableToolCount)
auto-extends coverage to cover every new tool: happy-path round-
trip with fence-shape assertion, 5xx error-path with MCP_ERROR fence
propagation, and 'every registered tool is dispatchable' guard.
- docs/reference/mcp.md 'Available Tools' table expanded from 16 to
22 resource domains with current per-domain tool counts.
Acceptance gate (verified):
- go build ./cmd/server/... ./cmd/agent/... ./cmd/cli/... ./cmd/mcp-server/...
clean across all four production binaries.
- go vet ./... clean.
- go test -short -count=1 ./internal/mcp/... pass (TestMCP_AllTools_*
expanded to 127 tool round-trips).
- go test -short -count=1 ./... pass repo-wide.
- bash scripts/ci-guards/openapi-handler-parity.sh clean (router 178,
OpenAPI 144, exceptions 36 — unchanged; we add MCP wrappers, not
routes).
- gofmt -l clean across the four touched files.
|
||
|
shankar0123
|
25c34ace45 |
feat(mcp): add claim_discovered + dismiss_discovered MCP tools (I-2 closure)
Closes the LAST P1 in the 2026-04-24 audit (cat-i-b0924b6675f8). Pre-I-2
the README claimed "all API endpoints are exposed via MCP" but the
discovered-certificate lifecycle (HTTP handlers ClaimDiscovered +
DismissDiscovered at internal/api/handler/discovery.go:125,162) had
zero MCP tool wrappers — operators using Claude / Cursor / similar
MCP clients had no path to bring an out-of-band cert under management
or to mark a benign discovery as not-of-interest without dropping to
the REST API directly. The audit's count of 0 MCP discovery tools
was correct: `grep -niE 'discover|claim|dismiss' internal/mcp/tools.go`
returned only the pre-existing agent-retire tool's description text
mentioning sentinel discovery agents — no actual discovery-tool
registrations.
Added in internal/mcp/types.go:
- ClaimDiscoveredCertificateInput (id + managed_certificate_id)
- DismissDiscoveredCertificateInput (id)
Both follow the existing Go-doc / staticcheck convention (lead with
the type name + brief; closure-rationale prose follows). Pinned by
the existing L-1 staticcheck-fix lesson.
Added in internal/mcp/tools.go (slotted at end of file, after
certctl_auth_check):
- certctl_claim_discovered_certificate — POST /api/v1/discovered-certificates/{id}/claim
- certctl_dismiss_discovered_certificate — POST /api/v1/discovered-certificates/{id}/dismiss
Both wrap the existing HTTP handlers via the generic c.Post helper.
No backend changes; no openapi.yaml changes (both ops were already
in the spec from earlier work).
The audit's third name "acknowledge" is NOT closed: at recon, no
notification-acknowledge HTTP handler exists in the API surface
(grep across internal/api/handler/ returned zero hits for
"acknowledge"). The audit appears to have mis-quoted; "acknowledge"
isn't a real backend endpoint to wrap. If a future feature adds
notification acknowledgement, register it in the same shape.
Verification:
- go build ./... — clean
- go vet ./internal/mcp/... — clean
- go test ./internal/mcp/... -count=1 — pass
- golangci-lint v2.11.4 run ./... — 0 issues
- MCP tool count went from 85 → 87 (verify via `grep -cE 'gomcp\.AddTool\(' internal/mcp/tools.go`)
- S-1 + G-3 + D-1 + D-2 + B-1 + L-1 CI guardrails all still pass
Audit findings closed:
- cat-i-b0924b6675f8 (P1, MCP discovery completeness — last P1 in audit)
This brings the audit to ZERO REMAINING P1s.
Deferred follow-ups:
- Notification acknowledge MCP tool — add when a notification-ack
HTTP handler exists. Currently no such handler exists in the
API surface; treat as a separate feature, not an MCP gap.
|
||
|
shankar0123
|
2edac7e78b |
fix(mcp): close staticcheck ST1021 on BulkRenew/BulkReassign input docstrings
CI on the B-1 merge (
|
||
|
shankar0123
|
f0865bb051 |
fix(api,web,mcp): add bulk-renew + bulk-reassign endpoints, drop client-side N×HTTP loops (L-1 master)
Two audit findings, both category cat-l, both rooted in
web/src/pages/CertificatesPage.tsx. Pre-L-1 the GUI looped per-cert
HTTP calls — 100 selected certs = 100 sequential round-trips × ~50–200
ms each = a 5–20-second wedge during which the operator stared at a
progress bar. Post-L-1 each workflow is a single POST.
cat-l-fa0c1ac07ab5 [P1, primary] — bulk renew loop
handleBulkRenewal: for/await triggerRenewal(id)
cat-l-8a1fb258a38a [P2] — bulk reassign loop
handleReassign: for/await updateCertificate(id, {owner_id})
The bulk-revoke endpoint (POST /api/v1/certificates/bulk-revoke +
BulkRevocationCriteria/Result) already existed as the canonical shape
in v2.0.x — L-1 ports that pattern to renew + reassign with per-action
twists.
Backend (Go)
- internal/domain/bulk_renewal.go: BulkRenewalCriteria mirrors
BulkRevocationCriteria (criteria + IDs modes); BulkRenewalResult
envelope adds EnqueuedJobs[] for per-cert {certificate_id, job_id};
shared BulkOperationError type for all bulk paths.
- internal/domain/bulk_reassignment.go: narrower shape — IDs-only,
owner_id required, team_id optional.
- internal/service/bulk_renewal.go::BulkRenewalService.BulkRenew:
resolves criteria → status filter (Archived/Revoked/Expired/
RenewalInProgress all silent-skip) → per-cert status flip + job
create. Keygen-mode-aware so jobs land in the same initial status
as single-cert TriggerRenewal. Single bulk audit event per call,
not N.
- internal/service/bulk_reassignment.go::BulkReassignmentService.
BulkReassign: validates owner_id upfront via the
ErrBulkReassignOwnerNotFound typed sentinel — non-existent owner
returns 400 before any cert is touched. Already-owned-by-target
is silent-skip. Single bulk audit event.
- internal/api/handler/{bulk_renewal,bulk_reassignment}.go: HTTP
shape mirrors bulk_revocation.go. NOT admin-gated (renew is non-
destructive; reassign is a common-case workflow). Sentinel-error
→ 400 mapping for OwnerNotFound.
- internal/api/router/router.go: three bulk-* routes registered as a
block before the {id} routes. HandlerRegistry gains BulkRenewal +
BulkReassignment fields.
- cmd/server/main.go: NewBulkRenewalService threads cfg.Keygen.Mode
so bulk-renew jobs land in same initial state as single-cert path.
Frontend
- web/src/api/client.ts: bulkRenewCertificates(criteria) +
bulkReassignCertificates(request) functions with full TS types.
- web/src/pages/CertificatesPage.tsx: handleBulkRenewal + handleReassign
rewritten from N-call loops to single calls. Result envelope drives
progress UI; first-error message surfaced when total_failed > 0.
Stale triggerRenewal + updateCertificate imports removed.
MCP
- internal/mcp/types.go: BulkRenewCertificatesInput +
BulkReassignCertificatesInput.
- internal/mcp/tools.go: certctl_bulk_renew_certificates +
certctl_bulk_reassign_certificates tools mirroring the existing
certctl_bulk_revoke_certificates pattern.
OpenAPI
- api/openapi.yaml: two new operations (bulkRenewCertificates,
bulkReassignCertificates) under Certificates tag. Four new schemas
(BulkRenewRequest, BulkRenewResult, BulkEnqueuedJob,
BulkReassignRequest, BulkReassignResult).
Tests
- Domain: BulkRenewalCriteria.IsEmpty + BulkReassignmentRequest.IsEmpty
IsEmpty contracts; JSON round-trip shape pinning.
- Service: 7 BulkRenew tests (happy/criteria-mode/skips-RenewalInProgress/
skips-revoked-archived/empty-criteria-error/partial-failure/
audit-event-emitted) + 8 BulkReassign tests (happy/skips-already-
owned/owner-required/empty-IDs/owner-not-found-sentinel/team-id-
optional/team-id-provided/partial-failure/audit-event-emitted).
- Handler: 5 BulkRenew handler tests (happy/empty-body-400/wrong-
method-405/actor-attribution/service-error-500) + 6 BulkReassign
handler tests (happy/empty-IDs-400/missing-owner-400/owner-not-
found-400-via-sentinel/wrong-method-405/generic-error-500).
CI guardrail
- .github/workflows/ci.yml: 'Forbidden client-side bulk-action loop
regression guard (L-1)'. Greps web/src/pages/CertificatesPage.tsx
for 'for(...) await triggerRenewal(...)' and 'for(...) await
updateCertificate(...)' patterns; comment lines exempt; test files
exempt. Verified locally (passes against post-fix tree, fires
against synthetic regression).
Counts (deltas)
- Routes: 119 → 121 (+2)
- OpenAPI operations: 123 → 125 (+2)
- MCP tools: 83 → 85 (+2)
Performance
- 100-cert bulk-renew: ~10s of sequential HTTP → ~100ms (99% latency
reduction on the canonical operator workflow).
- Audit event volume: 1 + N per operation → 1.
Out of scope (deferred follow-ups)
- cat-b-31ceb6aaa9f1: updateOwner/updateTeam/updateAgentGroup orphan
(different shape — wire existing PUT to GUI, not new bulk endpoint).
- cat-k-e85d1099b2d7: CertificatesPage no pagination UI.
- cat-i-b0924b6675f8: MCP missing claim/dismiss/acknowledge (L-1 added
2 new tools but does not close that finding).
Verification
- go build / vet / test -short / test -short -race all clean.
- web tsc --noEmit + vitest run all clean (296 tests passing).
- OpenAPI YAML parses (89 paths, 125 ops).
- L-1 CI guardrail passes against post-fix tree, fires against
synthetic regression.
No push.
|
||
|
shankar0123
|
675b87ba63 |
I-005: notification retry loop + dead-letter queue
Critical alerts can no longer be silently dropped by a transient
notifier failure. Failed notification attempts now ride an exponential
backoff retry loop, with a 5-attempt budget before promotion to the
dead-letter queue for operator intervention.
Schema (migration 000016, idempotent):
- retry_count INTEGER NOT NULL DEFAULT 0
- next_retry_at TIMESTAMPTZ
- last_error TEXT
- idx_notification_events_retry_sweep partial index
(next_retry_at) WHERE status='failed' AND next_retry_at IS NOT NULL
Dead rows clear next_retry_at so the index stops matching them.
Service contract:
- NotificationService.RetryFailedNotifications drives 2^n-minute
exponential backoff capped at 1h (notifRetryBackoffCap) with
5-attempt budget (notifRetryMaxAttempts).
- Exhaustion (RetryCount >= notifRetryMaxAttempts-1) promotes to
status='dead' via MarkAsDead.
- Non-terminal failures record via RecordFailedAttempt.
- Success path promotes to 'sent' without touching retry_count
(audit preserves "delivered on attempt N").
- Missing-notifier branch defensively promotes to 'sent' to avoid
wedging a row on a deleted channel.
- RequeueNotification operator escape hatch atomically resets
retry_count -> 0, next_retry_at -> NULL, last_error -> NULL,
status -> pending via notifRepo.Requeue.
Scheduler:
- New always-on notificationRetryLoop wired into the base loop set at
CERTCTL_NOTIFICATION_RETRY_INTERVAL (default 2m).
- sync/atomic.Bool idempotency guard.
- sync.WaitGroup shutdown drain via WaitForCompletion.
StatsService:
- SetNotifRepo setter pattern preserves 9 pre-existing
NewStatsService call sites (main.go + stats_test.go + 8 digest
tests) without touching the constructor signature.
- DashboardSummary.NotificationsDead populated via
notifRepo.CountByStatus(ctx, "dead") — nil-safe when unwired
(reports zero on systems without a notification repository).
- CountByStatus error is non-fatal (dashboard summary is
best-effort for this field).
- Prometheus certctl_notification_dead_total counter emitted from
the same snapshot.
Handler:
- New POST /api/v1/notifications/{id}/requeue endpoint.
- dead status surfaces to MCP + CLI.
Frontend:
- NotificationsPage gains two-tab toolbar ("All" / "Dead letter")
with queryKey: ['notifications', activeTab] so switching tabs
doesn't serve stale data until the 30s refetch.
- Dead rows surface "Retry {n}/5" + truncated last_error with
full-text title tooltip.
- Requeue mutation wrapped as
mutationFn: (id: string) => requeueNotification(id)
to prevent react-query v5's positional context argument from
leaking into the API client — pinned against future refactors
by strict-match toHaveBeenCalledWith('notif-dead-001') in
NotificationsPage.test.tsx:181.
Closes I-005.
|
||
|
shankar0123
|
0725713e19 |
Close I-004 (agent hard-delete cascades targets) coverage-gap finding
Operator decision answered as full soft-delete with optional forced
cascade — hard-delete is not reachable from any public surface. Prior
to this commit, DELETE /agents/{id} ran a plain `DELETE FROM agents`
whose schema-level `ON DELETE CASCADE` on deployment_targets.agent_id
silently wiped every target, orphaning certs and aborting in-flight
jobs. The finding closure reshapes the agent-removal contract around
soft retirement with explicit preflight counts, an opt-in cascade
gated by a mandatory reason, and unconditional protection for the
four reserved sentinel agents used by discovery sources.
Schema — migration 000015:
migrations/000015_agent_retire.up.sql flips
deployment_targets_agent_id_fkey from ON DELETE CASCADE to ON DELETE
RESTRICT, so a stray `DELETE FROM agents` now errors at the DB
boundary instead of quietly destroying targets. Both `agents` and
`deployment_targets` grow a retired_at TIMESTAMPTZ + retired_reason
TEXT pair (TEXT not VARCHAR so operator comments are never
truncated), indexed via partial indexes WHERE retired_at IS NOT
NULL. The migration is self-healing (ADD COLUMN IF NOT EXISTS, DROP
CONSTRAINT IF EXISTS then ADD CONSTRAINT, CREATE INDEX IF NOT
EXISTS) so repeated runs against partially-migrated databases
converge. migrations/000015_agent_retire.down.sql restores CASCADE
and drops the new columns for clean rollback. A dedicated
repository-layer testcontainers test
(internal/repository/postgres/migration_000015_test.go) asserts the
before/after FK action, column presence, index presence, and
round-trip idempotency under up→down→up.
Domain — sentinel guard + dependency counts:
internal/domain/connector.go gains IsRetired() on Agent, the
exported SentinelAgentIDs slice listing server-scanner,
cloud-aws-sm, cloud-azure-kv, cloud-gcp-sm verbatim (matching the
four reserved IDs documented in CLAUDE.md and created at startup in
cmd/server/main.go), IsSentinelAgent(id string) predicate,
AgentDependencyCounts{ActiveTargets, ActiveCertificates,
PendingJobs} with a HasDependencies() method, and ActorTypeAgent /
ActorTypeSystem enum values used by audit emission downstream.
Coverage locked down by internal/domain/connector_test.go.
Service — 8-step ordered contract:
internal/service/agent_retire.go:RetireAgent(ctx, id, actor,
opts{Force, Reason}) enforces a fixed execution order:
(1) sentinel guard — IsSentinelAgent(id) returns ErrAgentIsSentinel
unconditionally; force=true does NOT bypass it.
(2) fetch — ErrAgentNotFound on miss.
(3) idempotency — if IsRetired() already, return
AgentRetirementResult{AlreadyRetired: true} with no new audit
event and no state change (safe to replay from flaky clients).
(4) preflight counts — collectAgentDependencyCounts runs
ActiveTargets, ActiveCertificates, PendingJobs sequentially
(not in parallel; keeps the per-query timeout predictable and
matches the repo's existing call-chain shape).
(5) force-reason guard — opts.Force=true with empty Reason returns
ErrForceReasonRequired (wired into the 400 status surface).
(6) dependency guard — HasDependencies() with opts.Force=false
returns BlockedByDependenciesError{Counts} (wired into the 409
body with per-bucket counts).
(7) mutation — single pinned retiredAt := time.Now(); agent
retirement first, then cascade target retirement if opts.Force,
all under the repo's single transaction so the two retired_at
stamps match to the second.
(8) best-effort audit — agent_retired always; agent_retirement_
cascaded additionally on the force path. Actor is whatever the
handler resolves from the request; actor type is mapped by
resolveActorType (system/agent-prefix→Agent/else→User). Audit
emission failures are logged via slog.Error but do not abort
the retirement (matches the house convention used by every
other scheduler-emitted event).
BlockedByDependenciesError implements Error() as
"active_targets=%d, active_certificates=%d, pending_jobs=%d" and
Unwrap() → ErrBlockedByDependencies. The single struct satisfies
errors.Is via Unwrap (used by scheduler-level tests) and errors.As
via the concrete type (used by the handler to fish out Counts for
the 409 body). ListRetiredAgents(page, perPage) adds a separate
paginated accessor with page<1→1 and perPage<1→50 normalization so
retired rows are queryable without polluting the default agent
listing.
Sentinel guard coverage is asymmetric by design: all four reserved
IDs are protected, and force=true cannot override. Regression tests
in internal/service/agent_retire_test.go assert each of the eight
steps in order, plus sentinel bypass attempts and idempotency
replay.
Handler + router — status-code surface:
internal/api/handler/agents.go:RetireAgent exposes seven status
codes on DELETE /agents/{id}:
200 on a fresh retirement (body echoes AgentRetirementResult).
204 on idempotent replay (AlreadyRetired=true; no new audit).
400 on ErrForceReasonRequired.
403 on ErrAgentIsSentinel.
404 on ErrAgentNotFound.
409 on BlockedByDependenciesError, with a custom body shape
{error, counts{active_targets, active_certificates,
pending_jobs}} that bypasses the default ErrorWithRequestID
envelope so callers get the per-bucket numbers directly.
500 on any other error.
Heartbeat HandleHeartbeat returns 410 Gone when the agent is
retired (ErrAgentRetired), signalling the agent to shut down.
Query params `force=true` and `reason=<text>` drive the cascade
path; both are forwarded as url.Values through the new MCP
transport.
internal/api/router/router.go registers GET /api/v1/agents/retired
literal-path BEFORE /api/v1/agents/{id} — Go 1.22 ServeMux's
literal-beats-pattern-var precedence routes "retired" to the
paginated retired-agents listing instead of fetching a hypothetical
agent named "retired".
Agent binary — clean shutdown on 410:
cmd/agent/main.go gains the ErrAgentRetired sentinel, a
retiredOnce sync.Once, and a retiredSignal chan struct{}. A
markRetired(source, statusCode, body) helper closes the channel
exactly once; the Run() select loop observes the close and returns
ErrAgentRetired; main() matches via errors.Is(err, ErrAgentRetired)
and exits cleanly instead of spinning in the heartbeat retry loop.
The 410 Gone surface is therefore terminal for the agent process.
MCP transport:
internal/mcp/client.go adds Client.DeleteWithQuery(path, query),
a new additive transport method. Client.Delete is path-only; without
this method the retire tool would silently drop `force` and `reason`,
turning every cascade retire into a default soft-retire. The new
method shares do()'s 204 normalization and 4xx/5xx error
propagation so tool authors get one contract.
internal/mcp/tools.go + internal/mcp/types.go expose the
retire_agent tool with Force+Reason inputs wired through
DeleteWithQuery.
CLI:
cmd/cli/main.go + internal/cli/client.go add two CLI surfaces:
`agents list --retired` (client-side strip of --retired then
delegation to ListRetiredAgents, sharing --page/--per-page parsing
with the default listing) and `agents retire <id> [--force --reason
"…"]` (mirrors ErrForceReasonRequired — force without reason is
rejected client-side before the request is sent). JSON + table
output modes both honor the new columns.
Frontend:
web/src/pages/AgentsPage.tsx surfaces retired/retire affordances.
web/src/api/client.ts + web/src/api/types.ts expose the retire
endpoint and the retired-listing. 4 new Vitest regression cases.
OpenAPI:
api/openapi.yaml documents DELETE /agents/{id} with all seven
status codes, 410 on heartbeat, and the 409 per-bucket body shape.
Regression coverage (six new test files, all green):
internal/service/agent_retire_test.go — 8-step contract + sentinel guards
internal/api/handler/agent_retire_handler_test.go — 7-status-code surface + 410 heartbeat
internal/mcp/retire_agent_test.go — DeleteWithQuery wire-through
internal/cli/agent_retire_test.go — --retired listing + --force/--reason pairing
internal/repository/postgres/migration_000015_test.go — FK flip + columns + indexes + up↔down
internal/domain/connector_test.go — IsRetired, IsSentinelAgent, SentinelAgentIDs, HasDependencies
Files:
api/openapi.yaml — DELETE + 410 + 409 body shape
cmd/agent/main.go — ErrAgentRetired, markRetired, retiredSignal
cmd/cli/main.go — handleAgents list/get/retire dispatch
docs/architecture.md, docs/concepts.md,
docs/testing-guide.md — retirement contract narrative
internal/api/handler/agents.go — RetireAgent, status surface, 410 on heartbeat
internal/api/handler/agent_handler_test.go — extended coverage
internal/api/handler/agent_retire_handler_test.go — new
internal/api/router/router.go — /agents/retired before /agents/{id}
internal/cli/agent_retire_test.go — new
internal/cli/client.go — ListRetiredAgents + RetireAgent
internal/domain/connector.go — IsRetired, SentinelAgentIDs,
IsSentinelAgent, AgentDependencyCounts,
ActorTypeAgent/System
internal/domain/connector_test.go — new
internal/integration/lifecycle_test.go — retirement fixture
internal/mcp/client.go — DeleteWithQuery additive transport
internal/mcp/retire_agent_test.go — new
internal/mcp/tools.go, internal/mcp/types.go — retire_agent tool + Force/Reason inputs
internal/repository/interfaces.go — AgentRepository retirement methods
internal/repository/postgres/agent.go — retire + cascade target retire + counts
internal/repository/postgres/migration_000015_test.go — new
internal/service/agent.go — wire into AgentService surface
internal/service/agent_retire.go — new 8-step contract
internal/service/agent_retire_test.go — new
internal/service/deployment.go — skip retired agents
internal/service/target.go — skip retired agents
internal/service/testutil_test.go — shared mocks extended
migrations/000015_agent_retire.up.sql — new
migrations/000015_agent_retire.down.sql — new
web/src/api/client.ts, types.ts + tests — retire endpoint wiring
web/src/pages/AgentsPage.tsx — retire UI
|
||
|
shankar0123
|
a53a4b845b |
fix(gui,api): close C-001 + C-002 — ownership + agent FK contract
C-001 — CreateCertificate was server-accepted with null owner_id,
team_id, renewal_policy_id because the GUI neither collected the fields
nor enforced them, even though the backend's ManagedCertificate schema
and handler contract treat them as required. Fix the contract at all
four layers:
- web/src/pages/CertificatesPage.tsx: replace owner_id/team_id free-
text inputs with <select> elements fed by getOwners/getTeams/
getPolicies queries; mark all three required; gate the Create
button on owner_id + team_id + renewal_policy_id being set.
- internal/api/handler/certificates.go: ValidateRequired for
owner_id, team_id, renewal_policy_id on CreateCertificate so the
handler returns HTTP 400 with the offending field name before the
service layer is reached.
- internal/mcp/types.go: drop ',omitempty' from
CreateCertificateInput.RenewalPolicyID so the MCP schema reflects
the required contract; Update inputs keep partial-update semantics.
- api/openapi.yaml: 'required: [name, common_name, renewal_policy_id,
issuer_id, owner_id, team_id]' was already present on the Create
schema; clarified DeploymentTarget.agent_id description to note the
FK contract.
C-002 — CreateTargetWizard accepted an empty or bogus agent_id and the
service inserted directly, producing a Postgres 23503 FK-violation that
bubbled out as a generic HTTP 500. The FK itself (migration 000001 line
104: agent_id TEXT NOT NULL REFERENCES agents(id)) is correct; we keep
the schema strict and add validation at three layers:
- internal/service/target.go: introduce
ErrAgentNotFound sentinel and pre-validate agent_id in
TargetService.CreateTarget — empty string returns
'agent_id is required'; a nonexistent id returns the full
'referenced agent does not exist: <id>' error. Both wrap
ErrAgentNotFound via fmt.Errorf %w so callers can use errors.Is.
- internal/api/handler/targets.go: ValidateRequired on agent_id; map
errors.Is(err, service.ErrAgentNotFound) to HTTP 400 instead of
letting it fall through to the generic 500 branch.
- internal/mcp/types.go: drop ',omitempty' from
CreateTargetInput.AgentID to match the required contract.
- web/src/pages/TargetsPage.tsx: replace the free-text Agent ID input
with a <select> populated from getAgents(); include agent in the
canProceedToReview gate so Next is disabled until an agent is
chosen.
Regression coverage (21 new subtests total):
- TestCreateCertificate_MissingRequiredField_Returns400 — 6 subtests,
one per required field, each proves the handler guard fires before
the mock service is called.
- TestCreateTarget_MissingAgentID_Returns400 — handler guard.
- TestCreateTarget_NonexistentAgent_Returns400 — pins the
ErrAgentNotFound -> 400 translation.
- TestTargetService_CreateTarget_MissingAgentID — errors.Is sentinel.
- TestTargetService_CreateTarget_NonexistentAgentID — errors.Is.
- The existing TestTargetService_CreateTarget_Success, along with
TestCreateTarget_{MissingName,MissingType,NameTooLong}_* handler
tests, were updated to seed a real agent or include agent_id in
the request body so the happy paths still run cleanly.
Gates (Phase 4):
- go build/vet/test/race: green
- go test -cover: internal/service 68.7% (gate 55%),
internal/api/handler 78.9% (gate 60%)
- golangci-lint on service+handler+mcp: 0 issues
- govulncheck: no reachable vulns
- tsc --noEmit: clean
- vitest: 223/223 passing
See cowork/certctl-coverage-gap-audit.md entries C-001 and C-002.
|
||
|
shankar0123
|
eef1db0f0a |
fix(policies): stop 400ing the "+ New Policy" button + add per-rule severity (D-005, D-006)
Coverage Gap Audit findings D-005 (P0) + D-006 (P1) fixed together in a
single commit because they share the same root cause — policy CRUD sending
values the backend silently rejects — and splitting them would leave a
half-working UI between commits.
## D-005 (P0): PoliciesPage dropdown 400s every Create Policy
Root cause
----------
`web/src/pages/PoliciesPage.tsx` populated the Type `<select>` from a
hardcoded `['key_algorithm', 'ownership', 'allowed_issuers', ...]` array.
The backend's `internal/api/handler/validators.go::ValidatePolicyType`
enforces the TitleCase allowlist `AllowedIssuers`, `AllowedDomains`,
`RequiredMetadata`, `AllowedEnvironments`, `RenewalLeadTime` — defined in
`internal/domain/policy.go`. Every Create Policy request was rejected with
`400 invalid policy type`. The error surfaced only as a transient toast;
the modal closed anyway. Silent user-visible failure.
Fix
---
- `web/src/api/types.ts`: added `POLICY_TYPES` and `POLICY_SEVERITIES`
tuples with `as const` and narrowed `PolicyRule.type`, `.severity`, and
`PolicyViolation.severity` to the literal-union types. Dropdown is now
sourced from the tuple; casing drift becomes a compile error.
- `web/src/pages/PoliciesPage.tsx`: rekeyed `severityStyles` /
`severityDots` to the TitleCase values, added `humanize()` for display
(AllowedIssuers → "Allowed Issuers"), removed the `badge-neutral`
fallback that was papering over the mismatch.
- `web/src/api/types.test.ts` (new): pins both tuples exactly. If anyone
edits one side of the frontend/backend contract without the other, CI
fails with a clear assertion. Pure-TS vitest, no RTL dependency.
## D-006 (P1): `severity` field silently dropped on create/update
Root cause
----------
`PolicyRule` had no `Severity` field in `internal/domain/policy.go`. The
frontend has always sent `severity` on create/update, but Go's
`json.Decoder` (default settings, no `DisallowUnknownFields`) silently
dropped it. The value never reached PostgreSQL. Every rule rendered with
the same severity because there was no severity — just a display
computation downstream.
Fix: option (b), full-stack schema add (not delete-the-field)
-------------------------------------------------------------
- Migration `000013_policy_rule_severity` (up + down): adds
`severity VARCHAR(50) NOT NULL DEFAULT 'Warning'` to `policy_rules` with
CHECK constraint `severity IN ('Warning', 'Error', 'Critical')`. No
index — three-value column on a low-thousands-rows table, planner will
seq-scan regardless. PG 11+ metadata-only ADD COLUMN, safe on live data.
- `internal/domain/policy.go`: added `Severity PolicySeverity` field.
- `internal/repository/postgres/policy.go`: plumbed `severity` through
ListRules SELECT + Scan, GetRule SELECT + Scan, CreateRule INSERT,
UpdateRule UPDATE (4 queries).
- `internal/service/policy.go::UpdatePolicy`: if the client omits
severity on a PUT (zero-value empty string), fetch the existing rule
and preserve its severity. Without this, partial updates would trip the
NOT NULL CHECK and 500. Preserves pre-existing behavior for Name/Type
(out of scope).
- `internal/api/handler/policies.go::CreatePolicy`: default empty severity
to `'Warning'`, then validate via `ValidatePolicySeverity`. 400 with
clear message instead of 500 on CHECK violation. `UpdatePolicy`:
validates severity only when provided.
- `internal/mcp/types.go` + `internal/mcp/tools.go`: added optional
`severity` on the MCP `create_policy` / `update_policy` tool inputs so
LLM callers stay in sync with the wire contract.
- `api/openapi.yaml`: added `severity` to the `PolicyRule` schema with
the enum and default.
Acceptance criterion (user-defined)
-----------------------------------
"Create a rule with severity=Critical, reload the page, and still see
Critical — no silent drops." Verified end-to-end: frontend sends
`severity: "Critical"`, handler validates, service persists, DB stores,
GET returns, React renders the correct badge.
Seed data
---------
`migrations/seed.sql`: four demo rules now have differentiated severities
— `pr-require-owner` → Warning, `pr-allowed-environments` → Error,
`pr-max-certificate-lifetime` → Critical, `pr-min-renewal-window` →
Warning. The user called out that seeding all four at the same severity
makes the feature look decorative; differentiation demonstrates the
column carries real signal.
## Integration test fix (side effect of D-006)
`internal/integration/e2e_test.go::TestCrossResourceWorkflow/CreatePolicy`
was sending `"severity": "High"` — a value from the pre-audit severity
vocabulary that the new `ValidatePolicySeverity` correctly rejects with
400. Changed to `"Error"` (closest semantic match in the new TitleCase
allowlist). Only severity reference in the integration/ directory;
verified via grep.
## Out of scope, logged for follow-up (d/D-008)
Three policy-engine drift issues orthogonal to D-005 + D-006, explicitly
deferred per direction:
1. `migrations/seed.sql` policy_rules INSERTs use lowercase TYPE values
(`'ownership'`, `'environment'`, `'lifetime'`, `'renewal_window'`).
These are load-bearing on `internal/service/policy.go::evaluateRule`'s
`switch rule.Type` (which also uses the lowercase strings). Migrating
requires coordinated changes across seed + evaluation engine.
2. `migrations/seed_demo.sql:482-483` contains lowercase `'critical'`
severity — will now fail the new CHECK constraint. Separate fix.
3. `evaluateRule` hardcodes `Severity: domain.PolicySeverityWarning` on
emitted violations and ignores the configured `rule.Config`. The new
severity column is read correctly on the CRUD path but not yet
consulted during evaluation.
## Verification
Backend:
- `go build ./...` — clean
- `go vet ./...` — clean
- `go test -short ./...` — all packages green, including
`internal/service` (policy service), `internal/api/handler` (policy +
MCP handler tests), `internal/integration` (e2e_test.go after fix),
`internal/domain`, `internal/repository/postgres`.
Frontend:
- `tsc --noEmit` — clean
- `vitest run` — 223/223 passing (4 new assertions in types.test.ts)
- `vite build` — clean (only the pre-existing chunk-size warning)
|
||
|
shankar0123
|
13cd4d98ba |
feat(V2.2): bulk revocation — filter-based fleet-wide certificate revocation
Add POST /api/v1/certificates/bulk-revoke with filter criteria (profile_id, owner_id, agent_id, issuer_id, team_id, certificate_ids), partial-failure tolerance, and audit trail. Includes MCP tool, CLI command (certs bulk-revoke), server-side bulk modal in GUI replacing client-side sequential loop, OpenAPI spec, compliance mapping updates, and 21 new tests (12 service, 7 handler, 1 CLI, 1 frontend). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> |
||
|
shankar0123
|
43a03c168c |
fix: Go 1.25 upgrade, codebase audit fixes, MCP server tests
Upgrade from Go 1.22 to 1.25 (minimum for MCP SDK, actively supported). CI updated to match. Codebase audit fixes: - Local CA parseIP() now uses net.ParseIP — IP SANs no longer silently dropped - Nil pointer guards in agent.go GetWorkWithTargets for target/cert enrichment - MCP CreateCertificateInput marks owner_id/team_id as required - NGINX connector uses CombinedOutput() — captures diagnostic output on failure - Jobs handler validates JSON decode on rejection body — returns 400 on malformed - CRL/OCSP handlers propagate requestID for error tracing MCP server tests (26 tests): - client_test.go: HTTP client coverage (GET/POST/PUT/DELETE, auth, 204, errors, binary) - tools_test.go: tool registration, pagination, end-to-end flows with mock API Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> |
||
|
shankar0123
|
956230aec1 |
feat: M18a — MCP server exposing all 76 API endpoints as AI-native tools
Separate standalone binary (cmd/mcp-server/) using official MCP Go SDK (modelcontextprotocol/go-sdk v1.4.1) with stdio transport. Stateless HTTP proxy translates MCP tool calls to certctl REST API requests. 76 tools across 16 resource domains with typed input structs and jsonschema tags for automatic LLM-friendly schema generation. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> |