feat: self-update, reconnect, model catalogue, build toolchain, docs fixes (#642, #716, #733, #724, #747) - #788
Closed
cluster2600 wants to merge 31 commits into
Closed
Conversation
- uninstall.sh: check parent directory writability instead of file writability when deciding whether sudo is needed for removal. The previous check followed symlinks, so a symlink to a user-writable target in a root-owned directory would attempt unprivileged rm and fail with EACCES. - scripts/install.sh: replace NODE_MGR-based sudo heuristic with a direct writability check on the npm global prefix. The old approach always used sudo for nodesource installs, which changes PATH and breaks environments where the prefix is already user-writable.
…in process args (NVIDIA#579, NVIDIA#664) Add redactSecrets() to runner.js with patterns matching env assignments, nvapi- prefixes, GitHub PATs, and Bearer tokens. Apply to error messages in run() and runInteractive() so failed commands never leak credentials. Fix setupSpark() to pass NVIDIA_API_KEY via env option instead of interpolating into the command string (visible in ps aux). Fix walkthrough.sh to use tmux -e for env propagation instead of embedding the key in the shell command. Fix telegram-bridge.js to use SSH SendEnv instead of command-line export. 9 new tests covering redaction patterns and regression guards.
…VIDIA#693) RFC1918 addresses were blocked by the egress proxy because no policy entry existed for host.openshell.internal. Add a local-inference preset that whitelists ports 11434 (Ollama) and 8000 (vLLM) with binary restrictions, and auto-suggest it during onboard when a local inference provider is selected.
Policy presets without a binaries section cause 403 errors because the egress proxy has no approved binary list and denies all traffic. - Add binaries (openclaw, claude) to all 9 affected presets - Package manager presets additionally allow their tool binaries (pip for pypi, npm/node/yarn for npm, docker for docker) - Add validatePreset() that warns on missing binaries at apply time - Add 3 regression tests ensuring all presets have binaries - Fix installer HTTPS remediation test (shadow real nemoclaw binary) - Fix installer to print GitHub URL when binary not found
…IA#606, NVIDIA#719) Add config-io.js module providing atomic JSON config file writes (write-to-temp + rename) with proper EACCES error detection and user-facing remediation hints. Refactor credentials.js and registry.js to use the shared module instead of raw fs calls. When ~/.nemoclaw/ is owned by root or has wrong permissions, users now get a clear error message with fix commands instead of a cryptic EACCES stack trace.
) Replace all curl|bash and curl|sudo bash patterns with download-to-file and SHA-256 verification before execution. Adds download_and_verify() helper to install.sh, scripts/install.sh, and scripts/brev-setup.sh. - Ollama installer: download to temp file, verify SHA-256, then execute - NodeSource setup: download to temp file, verify SHA-256, then execute - nvm installer: refactored to use shared download_and_verify() helper - NEMOCLAW_SKIP_INTEGRITY=1 env var available for CI/air-gapped bypass - 7 new tests covering hash match, mismatch, skip, cleanup, and static analysis of all three installer scripts
…NVIDIA#684) Ports 8080 (gateway) and 18789 (dashboard) now configurable via NEMOCLAW_GATEWAY_PORT and NEMOCLAW_DASHBOARD_PORT env vars. When the default port is occupied, resolvePort() auto-selects an alternative (port+1 through port+9) and notifies the user. Error messages now suggest the env var override. Adds parsePortEnv() with range validation (1024–65535). 15 new tests for port parsing, resolution, and conflict handling.
Add buildCredentialEnv() helper and KNOWN_CREDENTIAL_KEYS to credentials.js as the single source of truth for credential retrieval. All integrations now use getCredential() consistently instead of raw process.env access. - nemoclaw.js deploy: uses buildCredentialEnv() instead of per-key manual reads (fixes GITHUB_TOKEN bypassing credentials.json) - onboard.js sandbox create: uses buildCredentialEnv() for env injection - onboard.js setupInference: passes NVIDIA_API_KEY via env indirection (_NEMOCLAW_CRED) instead of CLI args (no longer visible in ps/proc) - onboard.js policy detect: removes redundant || process.env fallbacks - telegram-bridge.js: uses getCredential() for both tokens - 13 new tests covering buildCredentialEnv, getCredential priority, and source-level checks for consistent credential patterns
…DIA#626) k3s-inside-Docker has broken DNS on all platforms — CoreDNS forwards to 127.0.0.11 (Docker embedded DNS) which is unreachable from k3s pods. Previously the fix only applied to Colima, leaving Docker Desktop and plain Docker users with DNS failures (getaddrinfo EAI_AGAIN). - shouldPatchCoredns() now returns true for colima, docker-desktop, docker - fix-coredns.sh accepts a runtime argument and resolves the Docker socket per-runtime (Colima socket, Docker Desktop socket, or default) - onboard.js passes detected runtime to fix-coredns.sh - 4 new regression tests for DNS upstream resolution on non-Colima runtimes
…nt (NVIDIA#731, NVIDIA#726) install.sh: add --ignore-scripts to npm link on both source-checkout and GitHub-clone paths to prevent husky prepare hook from invoking missing binaries (npm error 127). onboard.js: lsof suggestions in port-conflict messages now include sudo prefix so users on Linux can actually see root-owned listeners. preflight.js: lsof port detection falls back to sudo -n when non-root user gets empty output, improving port conflict diagnosis. Also adds .shellcheckrc to suppress pre-existing info-level shellcheck warnings (SC2059 for printf color codes, SC1091 for dynamic nvm source).
…tallation (NVIDIA#737) Both installers now run detect_placeholder_package() before npm link, checking npm root -g for a placeholder nemoclaw package (no bin.nemoclaw field) and removing it. Post-install verification confirms the binary responds to --version. Docs updated to use git+https URL instead of bare npm install -g nemoclaw. 3 new regression tests.
…IA#578) Pin every third-party and official action across all 6 CI workflow files to full 40-character commit SHAs with version comments for auditability. Pinned actions: - actions/checkout v4 (34e11487) - actions/setup-node v4 (49933ea5) - actions/setup-python v5 (a26af69b) - actions/upload-artifact v4 (ea165f8d) - astral-sh/setup-uv v4 (38f3f104) - rossjrw/pr-preview-action v1 (ffa7509e) Added 7 regression tests that scan all workflow files and fail if any action uses a tag ref instead of a SHA.
The Dockerfile generated openclaw.json at build time with the default model (nemotron-3-super-120b-a12b), but model selection happened AFTER sandbox creation. Users who chose Ollama or vLLM saw the wrong model in the OpenClaw TUI. Fix: extract selectInferenceProvider() from setupNim() and call it BEFORE createSandbox(). New patchDockerfileModel() helper replaces ARG NEMOCLAW_MODEL in the build context so the image is built with the correct model baked into openclaw.json. Onboard step reorder: select provider (step 3) → create sandbox (step 4).
…int (NVIDIA#710, NVIDIA#709) Move Ollama validation into Step 3 (provider selection) so users get immediate feedback instead of failing in Step 5 after a multi-minute sandbox build. - hasInstalledOllamaModels(): checks `ollama list` before proceeding; exits with remediation hint (`ollama pull`) if no models are installed - getOllamaBindAddressHint(): on Linux, shows OLLAMA_HOST=0.0.0.0 fix with both one-shot and systemd-persistent instructions - validateLocalProvider() now runs early on Linux in selectInferenceProvider() to catch container-unreachable Ollama before sandbox creation - 8 new regression tests
New logger module writes timestamped DEBUG messages to stderr when enabled via --verbose, --debug, LOG_LEVEL=debug, or NEMOCLAW_VERBOSE=1. All runner commands are logged with secrets redacted. Flags are stripped before dispatch so they don't interfere with subcommands.
…lowedOrigins (NVIDIA#739) - Add OPENCLAW_VERSION build arg so users can upgrade OpenClaw without editing the Dockerfile (override via NEMOCLAW_OPENCLAW_VERSION env var) - Expand trustedProxies to include RFC1918 private CIDRs (10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16) so X-Forwarded-For headers are trusted from SSH tunnels and reverse proxies on headless deployments - Add NEMOCLAW_EXTRA_ORIGINS build arg for additional CORS origins (comma-separated, override via NEMOCLAW_EXTRA_ORIGINS env var) - Add patchDockerfileVersion() and patchDockerfileExtraOrigins() helpers in onboard.js, called during sandbox creation - 9 new regression tests
…esolved.conf (NVIDIA#744) On DGX Spark (Ubuntu with systemd-resolved), /etc/resolv.conf points to 127.0.0.53 (stub resolver) which is loopback and gets filtered out. resolve_coredns_upstream() now falls back to /run/systemd/resolve/resolv.conf which contains the actual upstream nameservers.
…in nimStatus (NVIDIA#684, NVIDIA#713) NIM inference port (default 8000) now configurable via NEMOCLAW_NIM_PORT env var, matching the pattern established for gateway and dashboard ports. nimStatus() health check uses stored port instead of hardcoded 8000, fixing incorrect health reports when NIM runs on a non-default port. nimPort is stored in the sandbox registry for later status queries.
OpenShell may set the sandbox home directory to 0711 (drwx--x--x), which prevents agents from listing or reading files. Add explicit chmod 755 /sandbox in the Dockerfile and a runtime fix_home_permissions() function in nemoclaw-start.sh that detects and repairs restrictive modes (700, 710, 711) at startup.
…hot, and policies Blueprint Python code previously had 0% test coverage. Adds 134 pytest tests across 3 modules: runner (plan/apply/status/rollback/CLI), snapshot (create/restore/cutover/rollback/list), and policy presets (structural validation of all 10 YAML presets with parametrized checks for required fields, binaries, port ranges, and SPDX headers).
Add 25 new tests covering previously untested branches: - validatePreset: warning branch when binaries section missing - extractPresetEntries: null return for missing network_policies - parseCurrentPolicy: all 3 code paths (null, no separator, with separator) - applyPreset: sandbox name validation (6 rejection cases + missing preset) - getAppliedPresets: unknown sandbox returns empty array - getProviderSelectionConfig: vllm-local provider, unknown provider returns null - getOpenClawPrimaryModel: default model selection for ollama/nim/unknown providers
…ests Export parsePolicyPresetEnv, isSafeModelId, getNonInteractiveProvider, getNonInteractiveModel, and printDashboard from onboard.js for testing. Add two new test files: - onboard-helpers.test.js: 34 tests covering policy preset parsing, model ID validation, non-interactive env var handling, and dashboard output rendering - branch-coverage.test.js: 40 tests targeting uncovered branches across platform.js (WSL procVersion, plain docker), preflight.js (getConfiguredPorts error paths), local-inference.js (default switch cases, validateOllamaModel JSON error), inference-config.js (model defaults), config-io.js (ENOENT/ENOTDIR), runner.js (shellQuote)
…-openshell, logger 26 new tests targeting uncovered branches: - CLI: --version, -v, --verbose/--debug flag stripping (6 tests) - config-io: non-EACCES error rethrow paths, no-cause constructor (3 tests) - platform: findColimaDockerSocket null, isUnsupportedMacosRuntime (6 tests) - resolve-openshell: empty/relative/absolute commandV, home edge cases (7 tests) - logger: format specifier overflow, plain message without args (2 tests) Coverage improvements: logger.js → 100% branches, config-io → 85%, resolve-openshell → 85.71%, platform → 84.31%, bin/lib → 81.92%
…models (NVIDIA#733) activeModelEntries() previously returned only the onboarded model, preventing users from switching to other models via `openshell inference set`. Now returns the full 6-model catalog (added kimi-k2.5 and qwen3.5-397b) with the onboarded model prepended if it's custom. Also fixes printDashboard() calling nimStatus() without the configured NIM port — now reads nimPort from registry for correct health checks on non-default ports.
NVIDIA#738) README.md: CLI reference link used .md extension instead of .html, causing 404 on docs.nvidia.com. PR template: relative links to CONTRIBUTING.md resolved from .github/ instead of repo root. setup-spark.sh: exit message now explains what was configured (cgroup v2) and why re-onboarding may be needed; removed unused SCRIPT_DIR variable (shellcheck SC2034). 6 new regression tests validate link integrity and messaging.
New `nemoclaw update` command detects installation type (source checkout or global npm) and updates accordingly. `nemoclaw update --check` shows whether an update is available without installing.
…#724) WhatsApp channel (via @whiskeysockets/baileys/libsignal-node) and other native addons (@matrix-org/matrix-sdk-crypto-nodejs, protobufjs, koffi) fail to install on DGX Spark because node:22-slim lacks build tools. Add build-essential + python3-dev to both Dockerfiles, then purge after all npm installs complete to keep the image lean (~200MB savings). 7 new regression tests validate toolchain presence, install ordering, and post-install cleanup in both production and test Dockerfiles.
The Dockerfile's Python config generator used shell interpolation
('${VAR}') inside Python string literals to read build args. A crafted
NEMOCLAW_OPENCLAW_VERSION or NEMOCLAW_EXTRA_ORIGINS containing single
quotes could inject arbitrary Python code during docker build.
Defense in depth:
- Dockerfile: promote ARGs to ENV, read via os.environ.get() instead
of shell interpolation — eliminates the injection vector at its root
- onboard.js: isSafeVersion() and isSafeOriginsList() validate inputs
before patching the Dockerfile, rejecting quotes/backticks/semicolons
- 15 new regression tests verify injection attempts are blocked and
the Dockerfile uses the safe os.environ pattern
…jection
detectGpu() now accepts optional {runCapture, platform} deps parameter
for testability without changing its public API (no args = real system).
20 new tests cover all GPU detection branches: NVIDIA discrete (single/
multi GPU VRAM), DGX Spark GB10 (free -m valid/empty/throws/NaN), Apple
Silicon (VRAM GB/MB, sysctl fallback valid/throws/empty), and all
error/fallthrough paths (nvidia-smi empty/non-numeric/throws, no
chipset match, system_profiler empty/throws, non-darwin skip).
… repair (NVIDIA#716) On WSL2 and other Docker runtimes, restarting openshell-cluster breaks gateway/sandbox connectivity, previously forcing a full re-onboard. The new `nemoclaw reconnect` command repairs connectivity in seconds: - Detects unhealthy gateway and restarts with version-pinned image - Re-applies CoreDNS fix for Docker-based runtimes - Waits for sandbox pod to reach Ready state - Restarts dashboard port forwards - `--diagnose` flag shows connectivity state without repair New reconnect.js module with full DI for testability. 32 new tests.
…DIA#736) The Dockerfile's Python config generator hardcoded reasoning=False and maxTokens=4096 for all models. Nemotron 3 Super 120B supports reasoning (thinking tokens) and 8192 max output tokens — when misconfigured, OpenClaw fails to decode streamed responses containing reasoning tokens, causing "error decoding response body" in the TUI. - Dockerfile: add reasoning_models set and model_max_tokens dict; use model_entry() helper instead of hardcoded model config objects - TypeScript plugin: add reasoning flag to ModelProviderEntry interface and MODEL_CATALOG entries (Super/Ultra=true, others=false) - 6 new regression tests (3 Dockerfile + 3 TypeScript)
Contributor
|
Warning Rate limit exceeded
⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Run ID: ⛔ Files ignored due to path filters (1)
📒 Files selected for processing (77)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
Contributor
|
This repository limits contributors to 10 open pull requests. Please close or merge existing PRs before opening new ones. |
10 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
graph TD subgraph "nemoclaw update" A[detectInstallType] -->|source| B[git fetch+reset+link] A -->|global| C[npm install -g] end subgraph "nemoclaw reconnect" D[checkGateway] -->|unhealthy| E[restart] E --> F[checkSandbox] F -->|unhealthy| G[repairCoreDns + ports] end subgraph "Model catalogue" H[activeModelEntries] --> I[6 models incl. kimi-k2.5, qwen3.5] endnemoclaw update [--check]via git or npmnemoclaw reconnect [sandbox] [--diagnose]build-essential+python3-devfor aarch64 native addonsCloses #642, #716, #733, #724, #747, #738.
Test plan