feat(kb): platform hardening batch — coverage v2, box naming, MCP multi-key, test-session resilience, custom prompt fix

CHANGELOG.md

@@ -6,6 +6,18 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 
 ## [Unreleased]
 
+### Changed
+
+#### KB compile box: distinct image name + pod prefix (operations)
+
+Renamed the KB compile box so operators can tell it apart from chat agentboxes at a glance during production troubleshooting (previously every pod was `agentbox-<id>` and the compile box could only be isolated by label/image filter).
+
+- **Image rename**: `kbc-compile-box` → `siclaw-kbc-box` (Makefile build/push targets, helm `siclaw.compileBoxImage` derivation, Dockerfile/README/docs). `agentbox.compileBoxImage` explicit-override semantics are unchanged.
+- **Pod-name prefix**: compile boxes now spawn as `kbc-box-<id>` instead of `agentbox-<id>`, declared on the BoxProfile (`podNamePrefix`) for the `kb-compile` / `kb-compile-codex` profiles. Derived resources (cert Secret, hostname) follow the prefix. Chat agentboxes and the read-only `kb-test` box keep the `agentbox-` prefix.
+- **Upgrade behavior**: on the first compile spawn after upgrade, the spawner reaps any pod left under the old `agentbox-<id>` name for that agent (guarded to compile boxes only — a chat box under the same name is never touched), so the old and new pods do not coexist.
+- **Transition double-push**: `make push-kbc` republishes the same image digest under the legacy name `kbc-compile-box` for one release cycle so runtimes still pinned to the old name keep resolving. Remove the legacy tag/push next release.
+- **DevOps action required**: the registry replication rule (ap-southeast → cn-shanghai) must gain an entry for `siclaw-kbc-box` (the old `kbc-compile-box` was not in the rule and had to be pushed manually). Until the rule is added, push `siclaw-kbc-box` to cn-shanghai manually for production.
+
 ### Added
 
 #### Prometheus Observability Layer

CLAUDE.md

@@ -99,7 +99,7 @@ mTLS is **K8s mode only**. Do not add mTLS dependencies to local mode code paths
 | `src/core/prompt.ts` | **⚠️ REQUIRES HUMAN APPROVAL** | — | Describe intent and wait for OK before editing |
 | `src/memory/` | invariants.md §7, decisions.md ADR-005 | `npm test` | Requires embedding config; pi-agent only |
 | `Dockerfile.agentbox` | security.md §3-5 | `docker build` | Dual-user model; capability set; setgid kubectl |
-| `kbc/platform/pod/*` (KB compile box, Python) | kbc/platform/pod/README.md | `cd kbc/platform/pod && python test_compile_box.py` (needs `pip install claude-agent-sdk aiohttp`; CI: kbc-ci.yml, path-filtered) | Box↔runtime HTTP+SSE contract is shared with `src/gateway/capability/session-driver.ts` + `server.ts` (`/session` body, event vocabulary) and `agentbox/box-profile.ts` (allowedTools names, ANTHROPIC env forward) — change both sides together. Behavior changes need a `kbc-compile-box` image rebuild + redeploy (runtime env `SICLAW_COMPILE_BOX_IMAGE`); a rolled-out runtime does NOT pick up a new box image for existing live sessions |
+| `kbc/platform/pod/*` (KB compile box, Python) | kbc/platform/pod/README.md | `cd kbc/platform/pod && python test_compile_box.py` (needs `pip install claude-agent-sdk aiohttp`; CI: kbc-ci.yml, path-filtered) | Box↔runtime HTTP+SSE contract is shared with `src/gateway/capability/session-driver.ts` + `server.ts` (`/session` body, event vocabulary) and `agentbox/box-profile.ts` (allowedTools names, ANTHROPIC env forward) — change both sides together. Behavior changes need a `siclaw-kbc-box` image rebuild + redeploy (runtime env `SICLAW_COMPILE_BOX_IMAGE`); a rolled-out runtime does NOT pick up a new box image for existing live sessions |
 | `k8s/` or `helm/` | security.md §5, invariants.md §11 | `helm template` | mTLS K8s-only; container hardening |
 | `src/agentbox/resource-handlers.ts` | invariants.md §1,6 | `npm test` | `materialize()` safe in K8s, destructive in local mode |
 | `src/core/job-registry.ts` | tools.md §9 | `npm test` (`job-registry.test.ts`) | `claimNotification` is the single-fire dedup; a completion notice fires exactly once across the process-exit vs `job_stop` race |

Makefile

@@ -24,7 +24,11 @@ RUNTIME_IMAGE  = $(REGISTRY)/siclaw-runtime:$(TAG)
 AGENTBOX_IMAGE = $(REGISTRY)/siclaw-agentbox:$(TAG)
 PORTAL_IMAGE   = $(REGISTRY)/siclaw-portal:$(TAG)
 OCR_IMAGE      = $(REGISTRY)/siclaw-ocr:$(TAG)
-KBC_IMAGE      = $(REGISTRY)/kbc-compile-box:$(TAG)
+KBC_IMAGE      = $(REGISTRY)/siclaw-kbc-box:$(TAG)
+# Transition alias (remove next release): the KB compile box's former name. The
+# push-kbc target double-pushes this same digest so runtimes still pinned to the
+# old name keep resolving during the rename window.
+KBC_LEGACY_IMAGE = $(REGISTRY)/kbc-compile-box:$(TAG)
 
 # ── OCI labels injected into every image ──
 DOCKER_LABELS = \
@@ -87,7 +91,7 @@ docker-portal: ## Build portal image
 docker-ocr: ## Build OCR backend image
 	docker build -f Dockerfile.ocr $(DOCKER_LABELS) -t $(OCR_IMAGE) .
 
-docker-kbc: ## Build KB compile-box image (spawned per compile run; helm agentbox.compileBoxEnabled derives this tag)
+docker-kbc: ## Build KB compile-box image siclaw-kbc-box (spawned per compile run; helm agentbox.compileBoxEnabled derives this tag)
 	cd kbc && docker build -f platform/pod/Dockerfile $(DOCKER_LABELS) -t $(KBC_IMAGE) .
 
 push: push-runtime push-agentbox push-portal push-ocr push-kbc ## Push all images to registry
@@ -104,8 +108,14 @@ push-portal: ## Push portal image
 push-ocr: ## Push OCR backend image
 	docker push $(OCR_IMAGE)
 
-push-kbc: ## Push KB compile-box image
+push-kbc: ## Push KB compile-box image siclaw-kbc-box (+ transition legacy alias)
 	docker push $(KBC_IMAGE)
+	# Transition double-push (remove next release): republish the SAME digest under
+	# the legacy name kbc-compile-box so a runtime pinned to it during the rename
+	# window still resolves. Drop these two lines once all runtimes ship on the
+	# siclaw-kbc-box name.
+	docker tag $(KBC_IMAGE) $(KBC_LEGACY_IMAGE)
+	docker push $(KBC_LEGACY_IMAGE)
 
 # ==================== Test ====================
 ##@ Test
@@ -132,6 +142,7 @@ info: ## Print build variables
 	@echo "PORTAL:      $(PORTAL_IMAGE)"
 	@echo "OCR:         $(OCR_IMAGE)"
 	@echo "KBC:         $(KBC_IMAGE)"
+	@echo "KBC(legacy): $(KBC_LEGACY_IMAGE)"
 
 logs: ## View recent logs (all components)
 	@echo "=== Runtime ===" && \

docs/design/2026-07-kb-authoring-message-idempotency.md

@@ -35,4 +35,4 @@ turning the run checkpoint into an unbounded transcript.
 The field is additive and optional, so Sicore, the runtime, and the compile-box
 image may roll independently. Full protection is active after all three pieces
 are deployed; changing `compile_box.py` requires rebuilding the
-`kbc-compile-box` image.
+`siclaw-kbc-box` image.

helm/siclaw/templates/_helpers.tpl

@@ -82,7 +82,7 @@ Build agentbox image string — same registry/tag as gateway, different componen
 {{/*
 KB compile-box image (spawned per compile run by the runtime; NOT a helm-managed
 pod). Release-coupled by default: agentbox.compileBoxEnabled=true derives
-{registry}/kbc-compile-box:{image.tag} — the image ships with every release
+{registry}/siclaw-kbc-box:{image.tag} — the image ships with every release
 (`make docker` builds it). agentbox.compileBoxImage overrides the full string
 (hot-fix the compile brain independently of a release) and implies enabled.
 Empty result ⇒ KB stays dark (fail-closed).
@@ -93,9 +93,9 @@ Empty result ⇒ KB stays dark (fail-closed).
 {{- $ab.compileBoxImage -}}
 {{- else if $ab.compileBoxEnabled -}}
 {{- if .Values.image.registry -}}
-{{- printf "%s/kbc-compile-box:%s" .Values.image.registry .Values.image.tag -}}
+{{- printf "%s/siclaw-kbc-box:%s" .Values.image.registry .Values.image.tag -}}
 {{- else -}}
-{{- printf "kbc-compile-box:%s" .Values.image.tag -}}
+{{- printf "siclaw-kbc-box:%s" .Values.image.tag -}}
 {{- end -}}
 {{- end -}}
 {{- end }}

helm/siclaw/values.yaml

@@ -196,13 +196,13 @@ agentbox:
   debugImage: ""
   # KB compile capability (kb-compile / kb-test boxes; built from this repo's
   # kbc/, ships with every release via `make docker`). The RELEASE-COUPLED
-  # switch: true ⇒ the runtime spawns {image.registry}/kbc-compile-box:{image.tag}
+  # switch: true ⇒ the runtime spawns {image.registry}/siclaw-kbc-box:{image.tag}
   # — same registry/tag convention as every other component. false + no
   # override ⇒ KB stays dark (fail-closed; compile attempts error, chat
   # unaffected).
   compileBoxEnabled: false
   # Full-image override (implies enabled): hot-fix the compile brain
-  # independently of a release, e.g. "registry/.../kbc-compile-box:main-<sha>".
+  # independently of a release, e.g. "registry/.../siclaw-kbc-box:main-<sha>".
   compileBoxImage: ""
   # Node selector applied to every spawned AgentBox pod. Constrains pods to
   # nodes carrying all of these labels (e.g. { disktype: ssd, pool: agents }).

kbc/platform/pod/README.md

@@ -48,11 +48,11 @@ platform/pod/.venv/bin/python platform/pod/test_compile_box.py
 ## Run (container, production form)
 
 ```bash
-docker build -f platform/pod/Dockerfile -t kbc-compile-box .
+docker build -f platform/pod/Dockerfile -t siclaw-kbc-box .
 docker run --rm -p 3000:3000 \
   -e ANTHROPIC_BASE_URL=https://<massapi>/ \   # model goes through the company massapi (key injected on the proxy side)
   -v /tmp/wd:/work \
-  kbc-compile-box
+  siclaw-kbc-box
 # then:
 #   POST :3000/sources {"run_id":"r1","bundle_base64":"...","bundle_sha256":"..."}
 # or for Source Snapshot v2: