refactor: eliminate runner — all targets use direct mode

.github/workflows/images.yml

@@ -44,39 +44,3 @@ jobs:
           labels: ${{ steps.meta.outputs.labels }}
           cache-from: type=registry,ref=${{ env.IMAGE_BASE }}:sandbox-cache
           cache-to: ${{ (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository) && format('type=registry,ref={0}:sandbox-cache,mode=max', env.IMAGE_BASE) || '' }}
-
-  runner:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-      - uses: actions/setup-go@v5
-        with:
-          go-version-file: go.mod
-      - name: Build harness binary
-        run: |
-          VERSION=$(git describe --tags --always 2>/dev/null || echo dev)
-          CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build \
-            -ldflags "-s -w -X main.version=${VERSION}" \
-            -o build/runner/harness .
-      - uses: docker/setup-buildx-action@v3
-      - uses: docker/login-action@v3
-        with:
-          registry: ${{ env.REGISTRY }}
-          username: ${{ github.actor }}
-          password: ${{ secrets.GITHUB_TOKEN }}
-      - uses: docker/metadata-action@v5
-        id: meta
-        with:
-          images: ${{ env.IMAGE_BASE }}
-          tags: |
-            type=semver,pattern=runner-v{{version}}
-            type=sha,prefix=runner-,format=short
-      - uses: docker/build-push-action@v6
-        with:
-          context: build/runner
-          platforms: linux/amd64
-          push: ${{ github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository }}
-          tags: ${{ steps.meta.outputs.tags }}
-          labels: ${{ steps.meta.outputs.labels }}
-          cache-from: type=registry,ref=${{ env.IMAGE_BASE }}:runner-cache
-          cache-to: ${{ (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository) && format('type=registry,ref={0}:runner-cache,mode=max', env.IMAGE_BASE) || '' }}

.gitignore

@@ -16,8 +16,6 @@ __pycache__/
 # Build artifacts
 harness
 harness-openshell
-build/runner/harness
-build/runner/openshell
 infracluster/
 .claude/
 

AGENTS.md

@@ -73,13 +73,13 @@ Current workarounds and their upstream tracking:
 | Custom sandbox image | Adds mcp-atlassian and GWS CLI to community base | Upstreaming MCP integrations |
 | `CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1` | Vertex AI rejects `context_management` beta header | Anthropic/Google to align APIs |
 | Atlassian `JIRA_URL`/`JIRA_USERNAME` as `--config` material | Provider v2 config keys not injected as env vars yet | OpenShell roadmap |
-| In-cluster runner Job | OpenShell doesn't have a native K8s-triggered sandbox creation | Potential future CRD |
 
 Previously worked around, now resolved:
 
 | Was | Resolution |
 |-----|-----------|
 | GWS credential export/upload to sandbox | GWS is now a native provider (`google-workspace`). Gateway manages OAuth refresh via `oauth2-refresh-token` strategy + `request_body_credential_rewrite` on `oauth2.googleapis.com`. Sandbox gets a proxy-resolved placeholder. |
+| In-cluster runner Job | Eliminated — all targets now use direct mode. The gateway is accessible via external Route + mTLS, so sandboxes are created from the user's machine. |
 
 ## Validation
 

Makefile

@@ -20,11 +20,10 @@ VERSION       := $(shell git describe --tags --always 2>/dev/null || echo dev)
 LDFLAGS       := -s -w -X main.version=$(VERSION)
 
 DEV_SANDBOX_IMAGE  := $(REGISTRY):sandbox-$(VERSION)
-DEV_RUNNER_IMAGE   := $(REGISTRY):runner-$(VERSION)
 
-.PHONY: all cli cli-runner \
+.PHONY: all cli \
         vet lint test test-local test-kind test-remote test-all \
-        dev-sandbox dev-runner dev-push clean help
+        dev-sandbox dev-push clean help
 
 ## ── CLI ──────────────────────────────────────────────────────────────
 
@@ -36,13 +35,6 @@ cli:
 	CGO_ENABLED=0 go build -ldflags '$(LDFLAGS)' -o harness .
 	@echo "Built: ./harness ($(VERSION))"
 
-## ── Images ────────────────────────────────────────────────────────────
-
-## Cross-compile harness binary for the runner image
-cli-runner:
-	CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build -ldflags '$(LDFLAGS)' -o build/runner/harness .
-	@echo "Built: build/runner/harness ($(VERSION))"
-
 ## ── Lint targets ─────────────────────────────────────────────────────
 
 ## Run go vet
@@ -80,10 +72,10 @@ test-kind: cli
 	SANDBOX_IMAGE=$(DEV_SANDBOX_IMAGE) CONTAINER_CLI=$(CONTAINER_CLI) ./test/kind-lifecycle.sh $(if $(KEEP),--keep)
 
 ## Remote (OCP): requires KUBECONFIG set
-test-remote: cli dev-sandbox dev-runner
+test-remote: cli dev-sandbox
 	@test -n "$${KUBECONFIG}" || { echo "ERROR: Set KUBECONFIG for OCP (e.g. export KUBECONFIG=infracluster/kubeconfig)"; exit 1; }
 	@echo ""
-	SANDBOX_IMAGE=$(DEV_SANDBOX_IMAGE) RUNNER_IMAGE=$(DEV_RUNNER_IMAGE) ./test/test-flow.sh ocp
+	SANDBOX_IMAGE=$(DEV_SANDBOX_IMAGE) ./test/test-flow.sh ocp
 
 ## All: unit + local + kind + remote
 test-all: test test-local test-kind test-remote
@@ -95,27 +87,20 @@ dev-sandbox:
 	$(CONTAINER_CLI) build -t $(DEV_SANDBOX_IMAGE) sandbox/
 	@echo "Built: $(DEV_SANDBOX_IMAGE)"
 
-## Build dev runner image locally
-dev-runner: cli-runner
-	$(CONTAINER_CLI) build --platform $(PLATFORM) -t $(DEV_RUNNER_IMAGE) build/runner/
-	@echo "Built: $(DEV_RUNNER_IMAGE)"
-
-## Build and push dev images (sandbox: multi-arch, runner: amd64)
-dev-push: cli-runner
+## Build and push dev sandbox image (multi-arch)
+dev-push:
 	@$(CONTAINER_CLI) rmi --force $(DEV_SANDBOX_IMAGE) 2>/dev/null || true
 	@$(CONTAINER_CLI) manifest rm $(DEV_SANDBOX_IMAGE) 2>/dev/null || true
 	$(CONTAINER_CLI) build --platform linux/amd64 --manifest $(DEV_SANDBOX_IMAGE) sandbox/
 	$(CONTAINER_CLI) build --platform linux/arm64 --manifest $(DEV_SANDBOX_IMAGE) sandbox/
 	$(CONTAINER_CLI) manifest push $(DEV_SANDBOX_IMAGE)
-	$(CONTAINER_CLI) build --platform $(PLATFORM) -t $(DEV_RUNNER_IMAGE) build/runner/
-	$(CONTAINER_CLI) push $(DEV_RUNNER_IMAGE)
-	@echo "Pushed: $(DEV_SANDBOX_IMAGE) (multi-arch) $(DEV_RUNNER_IMAGE) (amd64)"
+	@echo "Pushed: $(DEV_SANDBOX_IMAGE) (multi-arch)"
 
 ## ── Convenience targets ───────────────────────────────────────────────
 
 ## Clean built binaries
 clean:
-	rm -f harness build/runner/harness build/runner/openshell
+	rm -f harness
 	@echo "Cleaned binaries"
 
 ## Show available targets

README.md

@@ -138,7 +138,6 @@ See the [OpenShell docs](https://github.com/NVIDIA/OpenShell) for the full secur
 | `gateways/*/gateway.toml` | Deployment target config: `local/` (Podman), `kind/`, `ocp/` (OpenShift) |
 | `openshell.toml` | Deployment-level overrides (enabled providers, inference model, chart version) |
 | `sandbox/Dockerfile` | Sandbox image: OpenShell base + MCP servers + CLI tools |
-| `build/runner/Dockerfile` | Runner image: harness binary for in-cluster sandbox creation |
 | `sandbox/policy.yaml` | Network egress rules applied to sandboxes |
 
 ## Commands

SPEC.md

@@ -60,8 +60,7 @@ Full flow: deploy gateway, register providers, render agent config, create sandb
 4. **Render payload** -- `env.sh` (resolved env vars), `run.sh` (entrypoint wrapper), `task.md` (if set).
 5. **Create sandbox** -- `openshell sandbox create` with `--upload` (payload) and the startup command. Retry up to 5 times for supervisor race conditions.
 
-Local: sandbox created directly via the openshell CLI on the user's machine.
-Remote: a runner Job is deployed to the cluster (`harness launch`), which creates the sandbox from inside the cluster with mTLS gateway access.
+Both local and remote targets create the sandbox directly from the user's machine. Remote gateways are accessed via an external Route endpoint with mTLS authentication.
 
 ### `harness create [--agent NAME] [-f FILE] [--name SANDBOX]`
 
@@ -99,10 +98,6 @@ Stop or start a sandbox without deleting it. When NAME is omitted and exactly on
 
 Tear down resources. At least one flag required.
 
-### `harness launch` (hidden)
-
-In-cluster command for the runner Job. Reads agent config from `/etc/openshell/sandbox/agent.yaml` (mounted ConfigMap), configures mTLS gateway, renders payload, creates sandbox, sets up environment. Not meant for direct user invocation.
-
 ## Config Files
 
 | File | Purpose |
@@ -113,32 +108,30 @@ In-cluster command for the runner Job. Reads agent config from `/etc/openshell/s
 | `gateways/*/gateway.toml` | Deployment target config with Helm, images, RBAC |
 | `openshell.toml` | Deployment-level overrides (enabled providers, inference model) |
 | `sandbox/Dockerfile` | Sandbox image: OpenShell base + MCP servers + CLI tools |
-| `build/runner/Dockerfile` | Runner image: harness binary + openshell CLI |
 | `sandbox/policy.yaml` | Network egress rules applied to sandboxes |
 
 ## Image Tags
 
-All images are published to `ghcr.io/robbycochran/harness-openshell`. CI never publishes floating tags (`:latest`, `:sandbox`, `:runner`); the bare `:sandbox` fallback below exists only for local `go build` binaries without version ldflags.
+All images are published to `ghcr.io/robbycochran/harness-openshell`. CI never publishes floating tags (`:latest`, `:sandbox`); the bare `:sandbox` fallback below exists only for local `go build` binaries without version ldflags.
 
-| Trigger | Sandbox | Runner |
-|---------|---------|--------|
-| Release `v0.1.2` | `:sandbox-v0.1.2` | `:runner-v0.1.2` |
-| Any push/PR | `:sandbox-<sha>` | `:runner-<sha>` |
+| Trigger | Sandbox |
+|---------|---------|
+| Release `v0.1.2` | `:sandbox-v0.1.2` |
+| Any push/PR | `:sandbox-<sha>` |
 
 The CLI resolves images from its embedded version (set via `-ldflags` at build time):
 
 - `v0.1.2` → `:sandbox-v0.1.2` (tagged release)
 - `v0.1.2-5-gabc1234` → `:sandbox-v0.1.2-5-gabc1234` (dev build, matches `make dev-sandbox`)
 - `dev` → `:sandbox` (bare `go build` without ldflags)
 
-`SANDBOX_IMAGE` and `RUNNER_IMAGE` env vars override the version-based resolution.
+`SANDBOX_IMAGE` env var overrides the version-based resolution.
 
 ## Environment Variables
 
 | Variable | Purpose |
 |----------|---------|
 | `SANDBOX_IMAGE` | Override sandbox image (dev/CI builds) |
-| `RUNNER_IMAGE` | Override runner image (dev/CI builds) |
 | `HARNESS_DIR` | Override harness directory detection |
 | `OPENSHELL_NAMESPACE` | Override K8s namespace (default: `openshell`) |
 | `OPENSHELL_CLI` | Override openshell binary path |
@@ -148,7 +141,7 @@ The CLI resolves images from its embedded version (set via `-ldflags` at build t
 | `CONFIG_TOML` / `PROVIDERS_TOML` | Override paths to `openshell.toml` / `providers.toml` (preflight) |
 | `KUBECONFIG` | K8s cluster config for remote targets |
 
-`GATEWAY_ENDPOINT` and `GATEWAY_NAME` are internal — set on the in-cluster runner Job, not by users.
+`GATEWAY_NAME` is internal — used by env override in gateway config, not typically set by users.
 
 ## Payload
 

TODO.md

@@ -10,9 +10,7 @@
 - Prerequisite: proto files stabilize (OpenShell is alpha)
 
 ### Image registry as gateway config vs env override
-- gateway.toml `[images]` section sets sandbox/runner image refs
-- `SANDBOX_IMAGE`/`RUNNER_IMAGE` env vars override config (for dev/CI)
-- Two sources of truth: gateway.toml hardcodes a registry, env vars override it
+- `SANDBOX_IMAGE` env var overrides the version-based image resolution (for dev/CI)
 - Consider: gateway.toml uses a `registry` field and images are relative to it
 
 ### registerProviders should filter by agent's provider list
@@ -49,14 +47,12 @@
 ## Testing
 
 ### Current coverage
-- Go unit tests across cmd/ (including launch.go) and all internal/ packages
+- Go unit tests across cmd/ and all internal/ packages
 - 29 bats preflight tests (run in CI via `.github/workflows/ci.yml`)
 - Integration: local + kind + OCP via `make test-all`
 
 ### Gaps
 - [ ] Integration test for `providers --force`
-- [ ] Unit test for the full `runLaunch` orchestration (currently only its
-      helpers — configureGateway, checkProviders, launchCreateSandbox — are tested)
 
 ## Release
 
@@ -66,8 +62,6 @@
 
 ## Deferred (post-0.1)
 
-- [ ] Rename K8s SA `openshell-launcher` -> `openshell-runner` (breaking for deployed OCP clusters)
-- [ ] Rename `LauncherSection` -> `RunnerSection` in gateway config TOML
 - [ ] Gateway-level LLM proxy/logging (gateway.toml `[proxy]` section)
 - [ ] Multi-agent workflow support (fleet.yaml / workflow.yaml)
 - [ ] `harness policy suggest` (DenialEvent stream -> policy proposals)

cmd/create.go

@@ -5,7 +5,6 @@ import (
 	"os"
 	"path/filepath"
 	"slices"
-	"strings"
 	"time"
 
 	"github.com/robbycochran/harness-openshell/internal/agent"
@@ -38,12 +37,11 @@ func NewCreateCmd(harnessDir, cli string) *cobra.Command {
 
 			gw := gateway.New(cli)
 
-			// 1. Check which gateway is active and whether it's local or remote.
+			// 1. Check which gateway is active.
 			activeGW, err := activeGatewayInfo(gw)
 			if err != nil {
 				return err
 			}
-			isLocal := strings.Contains(activeGW.Endpoint, "127.0.0.1")
 
 			status.Header("Gateway")
 			status.OKf("%s (%s)", activeGW.Name, activeGW.Endpoint)
@@ -102,21 +100,8 @@ func NewCreateCmd(harnessDir, cli string) *cobra.Command {
 				}
 			}
 
-			// 6. Determine whether the in-cluster runner is needed.
-			needsRunner := false
-			if !isLocal {
-				needsRunner = profileHasCustomProviders(providerNames, allProviders)
-			}
-
 			// 6. Deploy the sandbox
 			status.Header("Creating sandbox")
-			if needsRunner {
-				status.Info("Custom providers detected — using in-cluster runner")
-				gwCfg := loadGatewayConfigForActive(harnessDir, activeGW)
-				return createViaRunner(harnessDir, gwCfg, gw, agentName, name)
-			}
-
-			// Render payload and create directly
 			payloadDir, err := os.MkdirTemp("", "harness-payload-")
 			if err != nil {
 				return fmt.Errorf("creating payload dir: %w", err)
@@ -167,49 +152,4 @@ func activeGatewayInfo(gw gateway.Gateway) (*gateway.GatewayInfo, error) {
 	return nil, fmt.Errorf("no active gateway — deploy one first: harness deploy")
 }
 
-func profileHasCustomProviders(providerNames []string, allProviders []preflight.Provider) bool {
-	custom := make(map[string]bool)
-	for _, p := range allProviders {
-		if p.Type == "custom" {
-			custom[p.Name] = true
-		}
-	}
-	for _, name := range providerNames {
-		if custom[name] {
-			return true
-		}
-	}
-	return false
-}
-
-func loadGatewayConfigForActive(harnessDir string, active *gateway.GatewayInfo) *gateway.GatewayConfig {
-	if active != nil && active.Name != "" {
-		dir := filepath.Join(harnessDir, "gateways", active.Name)
-		if cfg, err := gateway.LoadConfig(dir); err == nil {
-			return cfg
-		}
-	}
-	gwDir := filepath.Join(harnessDir, "gateways")
-	entries, err := os.ReadDir(gwDir)
-	if err != nil {
-		return nil
-	}
-	for _, e := range entries {
-		if !e.IsDir() {
-			continue
-		}
-		cfg, err := gateway.LoadConfig(filepath.Join(gwDir, e.Name()))
-		if err == nil && !cfg.IsLocal() {
-			return cfg
-		}
-	}
-	return nil
-}
-
-func createViaRunner(harnessDir string, gwCfg *gateway.GatewayConfig, gw gateway.Gateway, agentPath, sandboxName string) error {
-	if gwCfg == nil {
-		return fmt.Errorf("no gateway config found for remote gateway — expected gateways/<name>/gateway.toml")
-	}
-	return upRemote(harnessDir, gwCfg, gw, agentPath, sandboxName)
-}
 

cmd/create_test.go

@@ -5,7 +5,6 @@ import (
 	"testing"
 
 	"github.com/robbycochran/harness-openshell/internal/gateway"
-	"github.com/robbycochran/harness-openshell/internal/preflight"
 )
 
 func TestActiveGatewayInfo_ListError(t *testing.T) {
@@ -36,23 +35,4 @@ func TestActiveGatewayInfo_RemoteGateway(t *testing.T) {
 	}
 }
 
-func TestProfileHasCustomProviders_NoCustom(t *testing.T) {
-	allProviders := []preflight.Provider{
-		{Name: "github", Type: "openshell"},
-		{Name: "vertex-local", Type: "openshell"},
-	}
-	if profileHasCustomProviders([]string{"github", "vertex-local"}, allProviders) {
-		t.Error("no custom providers, should return false")
-	}
-}
-
-func TestProfileHasCustomProviders_WithCustom(t *testing.T) {
-	allProviders := []preflight.Provider{
-		{Name: "github", Type: "openshell"},
-		{Name: "gws", Type: "custom"},
-	}
-	if !profileHasCustomProviders([]string{"github", "gws"}, allProviders) {
-		t.Error("gws is custom, should return true")
-	}
-}