.coderabbit.yaml

# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
language: "en-US"
early_access: false
reviews:
  profile: "chill"
  request_changes_workflow: false
  high_level_summary: true
  poem: true
  review_status: true
  review_details: false
  auto_review:
    enabled: true
    drafts: false
  path_instructions:
    - path: "docs/**"
      instructions: |
        You are an editorial reviewer for NemoClaw documentation.
        Review every docs change against the style guide below. Flag violations
        inline as review comments. Do not rewrite the content — point out the
        issue and cite the relevant rule so the author can fix it.

        ## Voice and Tone

        - Active voice required. Flag passive constructions.
          Bad: "A gateway is created by the CLI."
          Good: "The CLI creates a gateway."
        - Second person ("you") when addressing the reader.
        - Present tense. Flag future tense ("will") in descriptions of current behavior.
        - No hedge words: flag "simply," "just," "easily," "of course."

        ## LLM-Generated Patterns (Flag These)

        These are common in AI-generated text and erode trust with technical readers.
        Flag every occurrence:

        - Unnecessary bold on routine instructions ("This is a **critical** step"
          when nothing is actually critical). Bold is reserved for UI labels,
          parameter names, and genuine warnings.
        - Excessive em dashes. One per paragraph is fine; multiple per paragraph
          or em dashes used instead of commas/periods should be flagged.
        - Superlatives and marketing language ("powerful," "robust," "seamless,"
          "cutting-edge"). Say what it does, not how great it is.
        - Emoji in documentation prose.
        - Rhetorical questions ("Want to secure your agents? Look no further!").
          State the purpose directly.
        - Filler introductions ("In this section, we will explore..."). Start
          with the content.

        ## Formatting Rules

        - Every sentence must end with a period.
        - One sentence per line in source (makes diffs readable). Flag paragraphs
          where multiple sentences appear on the same line.
        - CLI commands, file paths, flags, parameter names, and values must use
          inline `code` formatting.
        - Copyable CLI code blocks must use language-specific tags such as
          `bash`, `sh`, or `powershell`, and must not include prompt markers
          such as `$`. Reserve `console` blocks for terminal transcripts that
          include prompts, output, or interactive sessions.
        - Do not flag `$$nemoclaw` in docs. It is the correct build-time
          variant placeholder for shared OpenClaw and Hermes command examples.
        - When a docs PR adds a command-line example that works across all
          NemoClaw agent aliases, comment if it uses a concrete alias such as
          `nemoclaw` or `nemohermes`; ask the author to use `$$nemoclaw`
          instead so generated OpenClaw and Hermes docs render the right
          command name.
        - Use Fern callout components (<Note>, <Tip>, <Warning>) for callouts
          in MDX pages, not bold text or blockquotes. MyST admonitions are
          acceptable only in legacy .md pages that have not migrated yet.
        - No nested admonitions.
        - Do not number section titles. Flag "Section 1: ...", "Step 3: ...", etc.
        - No colons in titles. Flag "Inference: Cloud and Local" — should be
          "Cloud and Local Inference."
        - Colons should only introduce a list. Flag colons used as general
          punctuation between clauses.

        ## Word List (Flag Incorrect Usage)

        | Correct        | Incorrect (flag these)                              |
        |----------------|------------------------------------------------------|
        | NVIDIA         | Nvidia, nvidia                                       |
        | NemoClaw       | nemoclaw (in prose), Nemoclaw                        |
        | OpenClaw       | openclaw (in prose), Openclaw                        |
        | OpenShell      | Open Shell, openShell, Openshell, openshell (in prose)|
        | CLI            | cli, Cli                                             |
        | API key        | api key, API Key                                     |
        | mTLS           | MTLS, mtls                                           |
        | YAML           | yaml (in prose), Yaml                                |
        | gateway        | Gateway (unless starting a sentence)                 |
        | sandbox        | Sandbox (unless starting a sentence)                 |

        Words inside code blocks or inline code spans are exempt from the word
        list (e.g., `nemoclaw onboard` is correct).

        ## Page Structure

        When reviewing new pages, verify:
        - SPDX license header is present after frontmatter.
        - Frontmatter includes title, description, keywords, topics, tags,
          content type, difficulty, audience, and status fields.
        - H1 heading matches the `title.page` frontmatter value.
        - Page starts with a one- or two-sentence introduction.
        - Sections use H2 and H3, each starting with an introductory sentence.
        - A "Next Steps" section at the bottom links to related pages.

        ## Severity

        - Word list and voice violations: flag as suggestions.
        - Missing SPDX header, broken cross-references, or incorrect code
          block language: flag as issues.
        - LLM-generated patterns (bold overuse, em dashes, superlatives,
          hedge words): flag as suggestions with the note "LLM pattern detected."
    - path: "**/*.md"
      instructions: |
        This rule applies to all Markdown files project-wide. For files
        under docs/, the stricter docs/** rules defined earlier in this
        file take precedence; this rule adds baseline checks for every
        other Markdown file (README, CONTRIBUTING, architecture docs, etc.).

        - NVIDIA must be all caps (not Nvidia, nvidia).
        - NemoClaw, OpenClaw, and OpenShell must use correct casing.
        - No emoji in technical prose.

    # ── E2E test recommendations ──────────────────────────────────
    # Maps sensitive file paths to the nightly E2E jobs that exercise them.
    # CodeRabbit surfaces these as review comments on PRs that touch a
    # matched path, so reviewers know which E2E jobs to run before merge.
    #
    # When adding a new E2E job to nightly-e2e.yaml, add a matching
    # path_instructions entry below. The cross-validation test in
    # test/validate-e2e-coverage.test.ts verifies consistency.

    - path: "scripts/nemoclaw-start.sh"
      instructions: &e2e-entrypoint |
        This file is a sandbox entrypoint script. Changes affect every
        sandbox boot and are invisible to unit tests (Landlock, non-root
        execution, process lifecycle).

        **E2E test recommendation:**
        - `sandbox-survival-e2e` — gateway restart recovery
        - `sandbox-operations-e2e` — process recovery after gateway kill
        - `cloud-e2e` — full onboard + cloud inference
        - `openclaw-slack-pairing-e2e` — gateway/connect-shell OpenClaw
          pairing state root parity for Slack

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=sandbox-survival-e2e,sandbox-operations-e2e,cloud-e2e,openclaw-slack-pairing-e2e
        ```

    - path: "scripts/lib/sandbox-init.sh"
      instructions: *e2e-entrypoint

    - path: "Dockerfile"
      instructions: &e2e-dockerfile |
        This file affects the sandbox container image. Layer ordering,
        permissions, and baked config changes are only testable with a
        real container build.

        **E2E test recommendation:**
        - `cloud-e2e` — full onboard + cloud inference
        - `sandbox-survival-e2e` — gateway restart recovery
        - `hermes-e2e` — Hermes agent onboard + inference
        - `rebuild-openclaw-e2e` — workspace state survives rebuild
        - `openclaw-tui-chat-correlation-e2e` — baked OpenClaw chat send
          patch preserves user/assistant turn correlation

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=cloud-e2e,sandbox-survival-e2e,hermes-e2e,rebuild-openclaw-e2e,openclaw-tui-chat-correlation-e2e
        ```

    - path: "Dockerfile.base"
      instructions: *e2e-dockerfile

    - path: "nemoclaw-blueprint/scripts/http-proxy-fix.js"
      instructions: |
        This file is the L7 proxy rewrite script. Changes affect all
        inference routing through the proxy. FORWARD-mode path needs
        manual validation until a dedicated forward-proxy-e2e exists.

        **E2E test recommendation:**
        - `cloud-e2e` — full inference through the proxy chain
        - `inference-routing-e2e` — credential isolation + error classification

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=cloud-e2e,inference-routing-e2e
        ```

    - path: "src/lib/onboard.ts"
      instructions: |
        This file contains core onboarding logic. Changes here affect
        the full sandbox creation and configuration flow.

        **E2E test recommendation:**
        - `cloud-e2e` — full onboard + cloud inference
        - `sandbox-operations-e2e` — multi-sandbox lifecycle
        - `rebuild-openclaw-e2e` — workspace state survives rebuild
        - `channels-stop-start-e2e` — channel disable/enable lifecycle across
          rebuild with cached Telegram credentials
        - `messaging-compatible-endpoint-e2e` — Telegram + compatible endpoint
          inference.local regression
        - `bedrock-runtime-compatible-anthropic-e2e` — Bedrock Runtime
          compatible Anthropic endpoint adapter boundary
        - `hermes-discord-e2e` — Hermes Discord config schema + placeholder
          isolation
        - `hermes-slack-e2e` — Hermes Slack policy + Python placeholder egress
        - `openshell-gateway-upgrade-e2e` — stale Linux Docker-driver gateway
          process restart after OpenShell upgrade
        - `issue-3600-gpu-proof-optional-e2e` — direct GPU verification remains
          optional when CUDA probing fails before sandbox creation

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=cloud-e2e,sandbox-operations-e2e,rebuild-openclaw-e2e,channels-stop-start-e2e,messaging-compatible-endpoint-e2e,bedrock-runtime-compatible-anthropic-e2e,hermes-discord-e2e,hermes-slack-e2e,openshell-gateway-upgrade-e2e,issue-3600-gpu-proof-optional-e2e
        ```

    - path: "src/lib/onboard/bedrock-runtime.ts"
      instructions: &e2e-bedrock-runtime-compatible-anthropic |
        This file is part of the Bedrock Runtime compatible Anthropic endpoint
        path. Changes affect endpoint classification, hidden adapter startup,
        adapter-token isolation, and the sandbox-facing inference.local route.

        **E2E test recommendation:**
        - `bedrock-runtime-compatible-anthropic-e2e` — fake Bedrock Runtime
          endpoint through OpenClaw and Hermes with sandbox leak checks

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=bedrock-runtime-compatible-anthropic-e2e
        ```

    - path: "src/lib/inference/bedrock-runtime*.ts"
      instructions: *e2e-bedrock-runtime-compatible-anthropic
    - path: "src/lib/onboard/channel-state.ts"
      instructions: &e2e-channel-stop-start |
        This file controls disabled channel resolution used during onboard
        and rebuild. Changes affect whether `channels stop` persists across
        sandbox destroy/recreate and whether `channels start` reattaches
        cached credentials.

        **E2E test recommendation:**
        - `channels-stop-start-e2e` — Telegram channel stop/start lifecycle
          across rebuild with cached credentials

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=channels-stop-start-e2e
        ```

    - path: "src/lib/state/onboard-session.ts"
      instructions: *e2e-channel-stop-start

    - path: "src/lib/actions/sandbox/rebuild.ts"
      instructions: *e2e-channel-stop-start

    - path: "src/commands/sandbox/channels/**"
      instructions: *e2e-channel-stop-start

    - path: "src/lib/sandbox/channels-command-support.ts"
      instructions: *e2e-channel-stop-start

    - path: "test/e2e/test-channels-stop-start.sh"
      instructions: *e2e-channel-stop-start

    - path: "src/lib/actions/inference-set.ts"
      instructions: |
        This file switches the OpenShell inference route and patches the
        selected running agent config. Changes here affect OpenClaw and
        Hermes model switching without a sandbox rebuild.

        **E2E test recommendation:**
        - `openclaw-inference-switch-e2e` — OpenClaw route + config patch
          + live OpenClaw request after switch
        - `hermes-inference-switch-e2e` — Hermes route + config.yaml patch
          + live Hermes request after switch

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=openclaw-inference-switch-e2e,hermes-inference-switch-e2e
        ```

    - path: "src/nemoclaw.ts"
      instructions: |
        This file contains CLI dispatch, status, recovery, and connect
        functions. Changes affect sandbox lifecycle commands.

        **E2E test recommendation:**
        - `sandbox-survival-e2e` — gateway restart recovery
        - `sandbox-operations-e2e` — process recovery after gateway kill

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=sandbox-survival-e2e,sandbox-operations-e2e
        ```

    - path: "src/lib/cluster-image-patch.ts"
      instructions: &e2e-overlayfs |
        This file handles Docker 26+ overlayfs compatibility. Changes
        are only testable under real Docker + K3s execution.

        **E2E test recommendation:**
        - `overlayfs-autofix-e2e` — Docker 26+ nested-mount auto-fix

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=overlayfs-autofix-e2e
        ```

    - path: "src/lib/onboard/preflight.ts"
      instructions: *e2e-overlayfs

    - path: "src/lib/deploy/**"
      instructions: |
        This file contains deployment lifecycle logic (start/stop
        cloudflared tunnel).

        **E2E test recommendation:**
        - `tunnel-lifecycle-e2e` — start/stop cloudflared tunnel
        - `state-backup-restore-e2e` — backup/restore workspace state

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=tunnel-lifecycle-e2e,state-backup-restore-e2e
        ```

    - path: "src/lib/state/sandbox.ts"
      instructions: |
        This file manages sandbox state (backup, restore, rebuild,
        snapshot). Changes affect data persistence across sandbox
        lifecycle operations.

        **E2E test recommendation:**
        - `state-backup-restore-e2e` — backup/restore workspace state
        - `snapshot-commands-e2e` — snapshot create/list/restore lifecycle
        - `rebuild-openclaw-e2e` — workspace state survives rebuild

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=state-backup-restore-e2e,snapshot-commands-e2e,rebuild-openclaw-e2e
        ```

    - path: "src/lib/shields/**"
      instructions: |
        These files control shields down/up, config mutability, audit
        trail, and auto-restore timer.

        **Test recommendation:**
        - `test/vm-driver-privileged-exec-routing.test.ts` — CLI Vitest
          regression for direct Docker/VM privileged exec routing used by
          shields/config mutations
        - `shields-config-e2e` — live shields lifecycle + config get/set/rotate

        To run the live E2E selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=shields-config-e2e
        ```

    - path: "src/lib/sandbox/privileged-exec.ts"
      instructions: |
        This file selects the privileged mutation transport for host-side
        sandbox writes. Changes can break shields up/down and config set on
        Docker Desktop VM-driver sandboxes.

        **Test recommendation:**
        - `test/vm-driver-privileged-exec-routing.test.ts` — CLI Vitest
          coverage for VM/Docker direct-container routing and prefix
          collision handling
        - `shields-config-e2e` — live shields lifecycle + config get/set/rotate

        To run the live E2E selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=shields-config-e2e
        ```

    - path: "src/lib/sandbox/config.ts"
      instructions: |
        This file implements host-side config get/set/rotate. Changes to
        config writes, hash recomputation, or privileged exec routing affect
        Docker/VM sandboxes.

        **Test recommendation:**
        - `test/vm-driver-privileged-exec-routing.test.ts` — CLI Vitest
          coverage for direct privileged exec routing used by host-side
          config mutation
        - `shields-config-e2e` — live config get/set/rotate while shields
          transitions are exercised

        To run the live E2E selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=shields-config-e2e
        ```

    - path: "agents/hermes/**"
      instructions: |
        This directory contains the Hermes agent. Changes affect
        multi-agent onboarding, health probes, and inference routing.

        **E2E test recommendation:**
        - `hermes-e2e` — Hermes onboard + health probe + live inference
        - `hermes-inference-switch-e2e` — Hermes route + config.yaml patch
          + live Hermes request after switch
        - `hermes-discord-e2e` — Hermes Discord config schema + placeholder
          isolation
        - `hermes-slack-e2e` — Hermes Slack policy + Python placeholder egress
        - `hermes-onboard-security-posture-e2e` — Hermes onboard under a
          non-root host security posture
        - `rebuild-hermes-e2e` — Hermes upgrade path
        - `rebuild-hermes-stale-base-e2e` — Hermes rebuild with stale cached base image

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=hermes-e2e,hermes-inference-switch-e2e,hermes-discord-e2e,hermes-slack-e2e,hermes-onboard-security-posture-e2e,rebuild-hermes-e2e,rebuild-hermes-stale-base-e2e
        ```

    - path: "test/e2e/test-hermes-slack-e2e.sh"
      instructions: |
        This script validates Hermes Slack onboarding, policy scoping, Slack
        credential providers, and Python Slack API egress through OpenShell
        placeholder substitution.

        **E2E test recommendation:**
        - `hermes-slack-e2e` — Hermes Slack policy + Python placeholder egress

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=hermes-slack-e2e
        ```

    - path: "scripts/patch-openclaw-chat-send.js"
      instructions: &e2e-openclaw-tui-chat-correlation |
        This helper patches OpenClaw chat send behavior inside the sandbox
        image. Changes can affect TUI message ordering, duplicate sends, and
        user/assistant turn correlation when OpenClaw is the active agent.

        **E2E test recommendation:**
        - `openclaw-tui-chat-correlation-e2e` - hermetic TUI chat proof for
          correlated user/assistant turns

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=openclaw-tui-chat-correlation-e2e
        ```

    - path: "test/e2e/test-openclaw-tui-chat-correlation.sh"
      instructions: *e2e-openclaw-tui-chat-correlation

    - path: "test/openclaw-tui-chat-correlation.test.ts"
      instructions: *e2e-openclaw-tui-chat-correlation

    - path: "test/e2e/test-issue-4434-tui-unreachable-inference.sh"
      instructions: &e2e-issue-4434-tui-unreachable-inference |
        This script is the privileged live repro for #4434. Changes affect
        the Linux-only NVIDIA endpoint firewall block, OpenClaw TUI error
        visibility, and active-spinner shutdown proof.

        **E2E test recommendation:**
        - `issue-4434-tui-unreachable-inference-e2e` - opt-in live proof
          that unreachable NVIDIA inference surfaces a visible TUI error and
          stops the active spinner

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=issue-4434-tui-unreachable-inference-e2e
        ```

    - path: "test/issue-4434-tui-unreachable-inference.test.ts"
      instructions: *e2e-issue-4434-tui-unreachable-inference

    - path: "test/e2e/test-openclaw-slack-pairing.sh"
      instructions: |
        This script validates OpenClaw Slack DM pairing across the gateway
        and connect-shell runtime contexts with a hermetic fake Slack Socket
        Mode event and fake chat.postMessage reply.

        **E2E test recommendation:**
        - `openclaw-slack-pairing-e2e` - Slack pairing request approval and
          allowFrom state root parity

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=openclaw-slack-pairing-e2e
        ```

    - path: "test/e2e/lib/fake-slack-api.cjs"
      instructions: |
        This helper backs hermetic Slack REST and Socket Mode tests. Changes
        can alter Slack credential rewrite assertions, fake Socket Mode
        envelopes, or chat.postMessage capture.

        **E2E test recommendation:**
        - `messaging-providers-e2e` - Slack REST provider placeholder rewrite
        - `openclaw-slack-pairing-e2e` - fake Slack Socket Mode pairing flow

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=messaging-providers-e2e,openclaw-slack-pairing-e2e
        ```

    - path: "test/e2e/lib/slack-api-proof.sh"
      instructions: |
        This helper applies hermetic Slack REST and Socket Mode policies for
        messaging E2E scripts. Changes can affect REST credential rewrite
        probes and native websocket policy coverage.

        **E2E test recommendation:**
        - `messaging-providers-e2e` - Slack REST provider placeholder rewrite
        - `openclaw-slack-pairing-e2e` - fake Slack Socket Mode pairing flow

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=messaging-providers-e2e,openclaw-slack-pairing-e2e
        ```

    - path: "nemoclaw-blueprint/policies/**"
      instructions: |
        This directory contains network policy definitions and presets.
        Changes affect sandbox egress rules and SSRF filtering.

        **E2E test recommendation:**
        - `network-policy-e2e` — deny-by-default, whitelist, hot-reload, SSRF

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=network-policy-e2e
        ```

    - path: "test/e2e/test-issue-2478-crash-loop-recovery.sh"
      instructions: |
        STAYS_IN_PR_UNTIL_SHIP — long-running soak test for the gateway
        recovery preload chain hardening (issue #2478). Removed in the
        same commit that drops the test file before merge.

        **E2E test recommendation:**
        - `issue-2478-crash-loop-recovery-e2e` — gateway recovery preload
          chain hardening soak

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=issue-2478-crash-loop-recovery-e2e
        ```

    - path: "test/e2e/lib/security-posture-assertions.sh"
      instructions: &e2e-security-posture |
        This helper asserts the Linux Docker-driver posture that protects
        static rc-file shims and runtime proxy-env configure guards under
        a non-root host user. Changes here affect both full onboard
        security posture lanes.

        **E2E test recommendation:**
        - `openclaw-onboard-security-posture-e2e` — OpenClaw onboard under a
          non-root host security posture
        - `hermes-onboard-security-posture-e2e` — Hermes onboard under a
          non-root host security posture

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=openclaw-onboard-security-posture-e2e,hermes-onboard-security-posture-e2e
        ```

    - path: "test/e2e/test-full-e2e.sh"
      instructions: |
        This script drives the full OpenClaw onboard path. Changes can
        alter the install, sandbox creation, health, inference, and
        security posture assertions used by the OpenClaw nightly lanes.

        **E2E test recommendation:**
        - `cloud-e2e` — full onboard + cloud inference
        - `openclaw-onboard-security-posture-e2e` — OpenClaw onboard under a
          non-root host security posture

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=cloud-e2e,openclaw-onboard-security-posture-e2e
        ```

    - path: "test/e2e/test-hermes-e2e.sh"
      instructions: |
        This script drives the full Hermes onboard path. Changes can alter
        install, sandbox creation, health, inference, and security posture
        assertions used by the Hermes nightly lanes.

        **E2E test recommendation:**
        - `hermes-e2e` — Hermes onboard + health probe + live inference
        - `hermes-onboard-security-posture-e2e` — Hermes onboard under a
          non-root host security posture

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=hermes-e2e,hermes-onboard-security-posture-e2e
        ```

    # ── Split cloud-experimental tests (#2644) ──────────────────
    - path: "test/e2e/test-cloud-onboard-e2e.sh"
      instructions: &e2e-cloud-onboard |
        This script tests the public installer flow, Landlock read-only
        enforcement, API key leak detection, and inference.local HTTPS.

        **E2E test recommendation:**
        - `cloud-onboard-e2e` — public installer + sandbox health + security

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=cloud-onboard-e2e
        ```

    - path: "test/e2e/e2e-cloud-experimental/checks/**"
      instructions: *e2e-cloud-onboard

    - path: "test/e2e/test-cloud-inference-e2e.sh"
      instructions: &e2e-cloud-inference |
        This script tests live chat via inference.local and validates
        the skill filesystem layout inside the sandbox.

        **E2E test recommendation:**
        - `cloud-inference-e2e` — live cloud inference + skill filesystem

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=cloud-inference-e2e
        ```

    - path: "test/e2e/test-cron-preflight-inference-local-e2e.sh"
      instructions: |
        This script onboards a sandbox against the managed cloud
        inference provider, then loads OpenClaw's cron isolated-agent
        preflight runtime directly from the in-sandbox dist and calls
        `preflightCronModelProvider` against the onboarded
        `inference.local` provider/model, asserting the result reports
        `status: "available"` and never `EAI_AGAIN` or the
        "local provider endpoint is not reachable" message.

        Exercises the Dockerfile fetch-guard patch that opts the cron
        model-provider preflight into trusted env-proxy mode so the
        managed inference hostname (`inference.local`) routes through
        the OpenShell L7 proxy instead of failing pinned DNS lookup.
        The scheduler boundary is intentionally bypassed: the cron CLI
        surfaces require `operator.admin` scope, which the in-sandbox
        auto-pair approval sweep deliberately omits.

        **E2E test recommendation:**
        - `cron-preflight-inference-local-e2e` — direct cron preflight
          runtime probe via OpenShell L7 proxy

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=cron-preflight-inference-local-e2e
        ```

    - path: "test/e2e/test-openclaw-inference-switch.sh"
      instructions: |
        This script validates OpenClaw model/provider switching with
        `nemoclaw inference set`, including OpenShell route state,
        /sandbox/.openclaw/openclaw.json, config hash recomputation, and
        live requests after the switch.

        **E2E test recommendation:**
        - `openclaw-inference-switch-e2e` — OpenClaw route + config patch
          + live OpenClaw request after switch

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=openclaw-inference-switch-e2e
        ```

    - path: "test/e2e/test-hermes-inference-switch.sh"
      instructions: |
        This script validates Hermes model/provider switching with
        `nemohermes inference set`, including OpenShell route state,
        /sandbox/.hermes/config.yaml, config hash recomputation, .env
        preservation, and live requests after the switch.

        **E2E test recommendation:**
        - `hermes-inference-switch-e2e` — Hermes route + config.yaml patch
          + live Hermes request after switch

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=hermes-inference-switch-e2e
        ```

    - path: "test/e2e/test-skill-agent-e2e.sh"
      instructions: &e2e-skill-agent |
        This script tests skill injection into the sandbox and verifies
        the agent reads the skill and returns a verification token.

        **E2E test recommendation:**
        - `skill-agent-e2e` — skill injection + agent verification

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=skill-agent-e2e
        ```

    - path: "test/e2e/e2e-cloud-experimental/features/**"
      instructions: *e2e-skill-agent

    - path: "test/e2e/test-docs-validation.sh"
      instructions: &e2e-docs-validation |
        This script validates CLI/docs parity (nemoclaw --help vs
        commands.md) and markdown link resolution.

        **E2E test recommendation:**
        - `docs-validation-e2e` — CLI/docs parity + link validation

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=docs-validation-e2e
        ```

    - path: "test/e2e/e2e-cloud-experimental/check-docs.sh"
      instructions: *e2e-docs-validation

    - path: "docs/reference/commands.md"
      instructions: *e2e-docs-validation

    - path: "test/e2e/test-messaging-compatible-endpoint.sh"
      instructions: |
        This script validates the Telegram + OpenAI-compatible endpoint
        regression path with a local mock endpoint. Changes here affect
        the hermetic coverage for managed inference.local routing from
        messaging-triggered agent turns.

        **E2E test recommendation:**
        - `messaging-compatible-endpoint-e2e` — Telegram + compatible
          endpoint inference.local routing

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=messaging-compatible-endpoint-e2e
        ```

    - path: "test/e2e/test-kimi-inference-compat.sh"
      instructions: &e2e-kimi-inference-compat |
        This script validates Kimi K2.6 managed inference.local
        compatibility with a hermetic OpenAI-compatible mock endpoint.
        Changes here affect OpenClaw tool-call splitting and trajectory
        proof for the Kimi compatibility path.

        **E2E test recommendation:**
        - `kimi-inference-compat-e2e` — Kimi managed inference.local
          compatibility and safe exec splitting

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=kimi-inference-compat-e2e
        ```

    - path: "nemoclaw-blueprint/openclaw-plugins/kimi-inference-compat/**"
      instructions: *e2e-kimi-inference-compat

    - path: "test/e2e/test-bedrock-runtime-compatible-anthropic.sh"
      instructions: *e2e-bedrock-runtime-compatible-anthropic

    - path: "test/e2e/test-openshell-gateway-upgrade.sh"
      instructions: |
        This script validates the old OpenShell install upgrade guard for
        Linux Docker-driver gateway processes.

        **E2E test recommendation:**
        - `openshell-gateway-upgrade-e2e` — stale gateway process restart after
          OpenShell upgrade

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=openshell-gateway-upgrade-e2e
        ```

    - path: "test/e2e/test-gateway-health-honest.sh"
      instructions: |
        This script is the coverage guard for #3111. It sabotages the
        openshell-gateway binary with a shim that crashes on startup and
        asserts onboard does NOT falsely log "Docker-driver gateway is
        healthy" and does exit non-zero. Exercises the NemoClaw-side
        false-positive (detached-zombie isPidAlive + metadata-only
        isGatewayHealthy) regardless of OpenShell packaging choices.

        **Regression E2E test recommendation:**
        - Run `regression-e2e.yaml` with the jobs input set to
          gateway-health-honest-e2e to validate false-positive health
          checks when the gateway binary crashes on startup (#3111).

        To run selectively:
        ```
        gh workflow run regression-e2e.yaml --ref <branch> -f 'jobs=gateway-health-honest-e2e'
        ```

    - path: "test/e2e/test-sessions-agents-cli.sh"
      instructions: &e2e-sessions-agents-cli |
        This script exercises the host-side `nemoclaw <name> sessions` and
        `nemoclaw <name> agents` subcommand groups end-to-end: gateway RPC
        envelope round-trip for `sessions reset` / `sessions delete`, plus
        passthrough for `sessions list` and `agents add` / `agents delete`.
        These flows are invisible to unit tests because the wrapper layer
        depends on `openshell sandbox exec` and a live OpenClaw gateway.

        **E2E test recommendation:**
        - `sessions-agents-cli-e2e` — sessions + agents CLI wrapper coverage

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=sessions-agents-cli-e2e
        ```

    - path: "src/lib/actions/sandbox/sessions/**"
      instructions: *e2e-sessions-agents-cli

    - path: "src/lib/actions/sandbox/agents/**"
      instructions: *e2e-sessions-agents-cli

    - path: "src/commands/sandbox/sessions.ts"
      instructions: *e2e-sessions-agents-cli

    - path: "src/commands/sandbox/sessions/**"
      instructions: *e2e-sessions-agents-cli

    - path: "src/commands/sandbox/agents.ts"
      instructions: *e2e-sessions-agents-cli

    - path: "src/commands/sandbox/agents/**"
      instructions: *e2e-sessions-agents-cli

    - path: "test/e2e/test-jetson-nvmap-gpu.sh"
      instructions: &e2e-jetson-nvmap |
        This script is the reporter-workflow E2E for #4231: on a Jetson Orin
        host it onboards with GPU, inspects the sandbox user's groups and
        /dev/nvmap, runs the cuInit(0) CUDA proof inside the sandbox, and
        asserts `nemoclaw status` reports `(CUDA verified)`. It validates that
        the Jetson Docker GPU recreate grants the sandbox user the host group
        owning /dev/nvmap so CUDA can initialize. Invisible to unit tests
        because it needs real Tegra device nodes and a live sandbox.

        **E2E test recommendation:**
        - `gpu-jetson-nvmap-e2e` — Jetson /dev/nvmap CUDA usability + status
          proof (Jetson-gated; needs a Jetson/Tegra GPU runner with
          `vars.JETSON_E2E_ENABLED=true`)

        To run selectively:
        ```
        gh workflow run nightly-e2e.yaml --ref <branch> -f jobs=gpu-jetson-nvmap-e2e
        ```

    - path: "src/lib/onboard/docker-gpu-patch.ts"
      instructions: *e2e-jetson-nvmap

    - path: ".github/workflows/nightly-e2e.yaml"
      instructions: |
        This is the nightly E2E workflow definition. Changes here affect
        which tests run and how they are triggered.

        If a new E2E job is added, verify corresponding
        `path_instructions` entries exist in `.coderabbit.yaml` for the
        source files it covers.

        If a job is renamed or removed, update the corresponding aggregate
        job `needs` lists (`notify-on-failure`, `report-to-pr`, and
        `scorecard`).
chat:
  auto_reply: true