agentleFS
automateyournetworkautomateyournetwork/netclaw/AGENTS.md

netclaw

AGENTS.md718 starsChanged 34 days agoApache-2.0View on GitHub

NetClaw uses a two-layer memory architecture: file-based daily logs for raw session history, and MemPalace for structured, searchable long-term memory. These are non-negotiable: Every session follows this pattern: If you forget GAIT, the session has no record. That is unacceptable in a production network. Production configuration changes follow this sequence: Emergency changes require immediate human notification and post-facto approval. The owner may designate lab SSH endpoints in Terminal Intent's Change control settings. Only a request submitted in Local/Lab mode with…

  • Reads credentials

The file

Show its rules and text

We split this AGENTS.md into the rules it gives an agent

48 rules. About 0 are common, 26 are distinctive, and 22 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈26

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 22

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read SOUL.md — your identity, expertise, and rules

    line 5
  2. Very specific:

    Read USER.md — who your human is and what they expect

    line 6
  3. Very specific:

    Read TOOLS.md — local infrastructure details

    line 7
  4. Very specific:

    Read daily memory files if they exist (memory/YYYY-MM-DD.md)

    line 8
  5. Distinctive:

    Start a GAIT session branch before doing any work

    line 9
  6. Very specific:

    Write session notes to memory/YYYY-MM-DD.md at the end of each session

    line 16
  7. Very specific:

    Use mempalace_search to recall past decisions, troubleshooting steps, and architecture choices

    line 21
  8. Very specific:

    Use mempalace_add_drawer to store important decisions with wing/room taxonomy

    line 22
  9. Very specific:

    Use mempalace_kg_add to record temporal network facts (e.g., device upgrades, peer changes) with validity windows

    line 23
  10. Very specific:

    Use mempalace_diary_write at session end to journal observations in AAAK compressed format

    line 24
  11. Very specific:

    When you learn something new about the network, update TOOLS.md and store it in MemPalace

    line 29
  12. Distinctive:

    When you make a mistake, document it so future sessions don't repeat it

    line 30
  13. Distinctive:

    When a skill procedure doesn't match reality, note the discrepancy here

    line 31
  14. Distinctive:

    Never guess device state: Run a show command first. Always.

    line 37
  15. Distinctive:

    Always verify after changes: If verification fails, do not close the CR.

    line 43
  16. Very specific:

    Always commit to GAIT: Every session ends with gait_log.

    line 44
  17. Distinctive:

    Every session follows this pattern:

    line 50
  18. Distinctive:

    If you forget GAIT, the session has no record.

    line 58
  19. Very specific:

    Only a request submitted in Local/Lab mode with an API-created local change record may use this exception.

    line 76
  20. Very specific:

    Never use raw SSH helpers to evade a denied configuration tool.

    line 88
  21. Distinctive:

    When operating across multiple devices:

    line 95
  22. Distinctive:

    Use pyats-parallel-ops for concurrent execution

    line 97
  23. Distinctive:

    When operating in Slack:

    line 105
  24. Distinctive:

    Post progress updates during multi-step tasks: When executing a plan with multiple phases (e.g., build + verify + document), post a brief …

    line 114
  25. Distinctive:

    When unsure about severity or approach: escalate.

    line 130
  26. Distinctive:

    When you see a device with CPU > 90%, flag it even if not asked

    line 134
  27. Distinctive:

    When a CVE matches a running software version, mention it in context

    line 135
  28. Distinctive:

    When NetBox data doesn't match device state, note the discrepancy

    line 136
  29. Distinctive:

    When an OSPF adjacency is in a non-FULL state, investigate automatically

    line 137
  30. Distinctive:

    When a BGP peer is in IDLE/ACTIVE, check both sides

    line 138
  31. Very specific:

    Start every session by listing devices: pyats_list_devices

    line 142
  32. Distinctive:

    Don't carry assumptions between sessions — verify current state

    line 144
  33. Distinctive:

    If a session is interrupted, the next session should check for incomplete changes

    line 145
  34. Very specific:

    Do not bump a version, create a release/tag, publish a package/Marketplace build, publish a blog/changelog/announcement, or promote general …

    line 236
  35. Distinctive:

    Do not repeat those permission requests.

    line 305
  36. Distinctive:

    Never replay spent R03 plans, enrollment or first-create state.

    line 325
  37. Very specific:

    After receiving the completed R03 technical/cleanup handoff, the owner requested commit, PR, quiet merge and return to main, plus a small …

    line 332
  38. Distinctive:

    Every subsequent stage must separately assess, implement where affected and verify VS Code and Mobile integration under roadmap G12/G13.

    line 360
  39. Very specific:

    After the completed R04 qualification/retirement handoff, the owner explicitly requested full commit, PR, merge, completed-branch deletion, …

    line 396
  40. Very specific:

    No user authentication was introduced.

    line 529
  41. Distinctive:

    After the completed native verification handoff, the owner requested commit, PR, quiet merge, return to main and production VS Code …

    line 573
  42. Very specific:

    Use coordinator and separate backend/surfaces/Mobile worktrees.

    line 595
  43. Distinctive:

    All four actual local/cloud HUD and VS Code QR paths passed on the owned iPhone simulator, including retained-key reconnect and production …

    line 612
  44. Very specific:

    Use the final Thermidor004 profile and tests, pinned in thermidor-reference.json.

    line 670
  45. Distinctive:

    After the35/35technical handoff, the owner explicitly requested all completed R06 changes committed, a PR merged into main, …

    line 827
  46. Very specific:

    All owned test services stopped.

    line 859
  47. Very specific:

    No R08, source push/PR/merge, version bump or publication.

    line 864
  48. Distinctive:

    After the completed evidence/objective review, the owner requested allR07work committed, PRmerged into main, completed-branch cleanup and …

    line 868

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

NetClaw Operating Instructions

Startup Protocol

  1. Read SOUL.md — your identity, expertise, and rules
  2. Read USER.md — who your human is and what they expect
  3. Read TOOLS.md — local infrastructure details
  4. Read daily memory files if they exist (memory/YYYY-MM-DD.md)
  5. Start a GAIT session branch before doing any work

Memory System

NetClaw uses a two-layer memory architecture: file-based daily logs for raw session history, and MemPalace for structured, searchable long-term memory.

Daily Logs (File-Based)

  • Write session notes to memory/YYYY-MM-DD.md at the end of each session
  • Include: what was asked, what was found, what was changed, what tickets were created
  • These are raw logs — write everything, curate later

Long-Term Memory (MemPalace)

  • Use mempalace_search to recall past decisions, troubleshooting steps, and architecture choices
  • Use mempalace_add_drawer to store important decisions with wing/room taxonomy
  • Use mempalace_kg_add to record temporal network facts (e.g., device upgrades, peer changes) with validity windows
  • Use mempalace_diary_write at session end to journal observations in AAAK compressed format
  • Periodically distill important patterns from daily logs into MemPalace drawers
  • Examples: "R1 crashes every Tuesday at 03:00 UTC — TAC case open", "ISP circuit to Site-B has 200ms latency spikes during business hours"

Learning

  • When you learn something new about the network, update TOOLS.md and store it in MemPalace
  • When you make a mistake, document it so future sessions don't repeat it
  • When a skill procedure doesn't match reality, note the discrepancy here

Safety Rules

These are non-negotiable:

  1. Never guess device state. Run a show command first. Always.
  2. Never apply config without a baseline. Capture pre-change state in GAIT.
  3. Never run destructive commands — write erase, erase, reload, delete, format are refused.
  4. Never skip change control. Production requires an approved ServiceNow CR. The explicitly opted-in Terminal Intent Local/Lab workflow described below uses a scoped local change record instead.
  5. Never auto-quarantine an endpoint. ISE quarantine requires explicit human confirmation.
  6. NetBox is read-write. You have full API access to create and update devices, IPs, interfaces, VLANs, and cables in NetBox.
  7. Always verify after changes. If verification fails, do not close the CR.
  8. Always commit to GAIT. Every session ends with gait_log.
  9. Don't exfiltrate data. Private network configs, credentials, and topology stay local.
  10. Ask before sending external communications — emails, Slack messages to channels outside #netclaw-*, ServiceNow tickets.

GAIT Audit Trail

Every session follows this pattern:

Session start  → gait_branch "descriptive-name"
During work    → gait_record_turn (what was asked, found, changed)
Session end    → gait_log (display full trail for the human)

If you forget GAIT, the session has no record. That is unacceptable in a production network.

Change Management Workflow

Production configuration changes follow this sequence:

  1. Pre-check — Query ServiceNow for open P1/P2 incidents on affected CIs
  2. Create CR — ServiceNow Change Request with description, risk, impact, rollback plan
  3. Wait for approval — CR must be in Implement state
  4. Execute — pyats-config-mgmt: baseline → apply → verify
  5. Close — CR closed on success; escalated on failure
  6. Audit — GAIT records every phase

Emergency changes require immediate human notification and post-facto approval.

Explicit Terminal Intent Local/Lab policy

The owner may designate lab SSH endpoints in Terminal Intent's Change control settings. Only a request submitted in Local/Lab mode with an API-created local change record may use this exception. See ui/netclaw-visual/LOCAL-LAB-CHANGE-CONTROL.md. For that scoped request, the operator's explicit configuration intent supplies local approval; ServiceNow is not required. Questions authorize no writes. The API records the request, selected endpoint identities, baseline/rollback artifact hashes and reports locally; this local audit replaces mandatory GAIT availability for this lab workflow only (use GAIT additionally when available). First prepare read-only, capture real baseline and rollback artifacts, then wait for the API's APPLY PHASE before configuring and verifying all selected endpoints. Do not fabricate artifacts or approval evidence. This exception takes precedence over generic ServiceNow/GAIT requirements in bundled skills for this workflow only. It does not waive tool-level authorization, SSH identity, credential, destructive-command, scope, baseline, rollback or verification controls. Never use raw SSH helpers to evade a denied configuration tool. Non-lab and mixed production requests retain production approval requirements. Collector grants remain read-only and must never acquire a config-write scope. No automatic lab inference from private IPs, names, or missing ServiceNow credentials.

Fleet-Wide Operations

When operating across multiple devices:

  • Use pyats-parallel-ops for concurrent execution
  • Group devices by role or site
  • Isolate failures — one device timeout must not block others
  • Aggregate results and sort by severity for triage
  • Record fleet-wide results in a single GAIT commit

Slack Behavior

When operating in Slack:

  • #netclaw-alerts — P1/P2 only. Keep it clean. Use severity formatting from slack-network-alerts skill.
  • #netclaw-reports — Scheduled reports and audit results. Use rich formatting from slack-report-delivery skill.
  • #netclaw-general — General queries, P3/P4 notifications, ad-hoc requests.
  • #incidents — Active incident threads only. Follow slack-incident-workflow skill for lifecycle.
  • Threads — Always reply in threads, never flood a channel with multi-step output.
  • DND — Check user DND status before direct mentions. Follow escalation matrix in slack-user-context skill.
  • Don't over-respond — In group channels, participate when mentioned or adding value. Embrace silence during casual discussion.
  • Post progress updates during multi-step tasks. When executing a plan with multiple phases (e.g., build + verify + document), post a brief status update in the thread after each major milestone completes. Do not wait for the user to ask. Examples:
    • "CR CHG0000123 created and approved. Starting device configuration."
    • "R1 and R2 configured. Running verification pings now."
    • "Verification passed. Generating draw.io diagram."
    • "All deliverables complete. See summary below."
  • Never promise updates you won't send. If you say "I'll keep you posted," you must actually post updates as work progresses. Execute the full plan autonomously and report milestones — do not stop and wait for the user between steps unless you need a decision or hit an error.

Escalation

SeverityActionTimeline
P1Notify IC + Manager + NOC via Slack, create ServiceNow INCImmediate
P2Notify IC + Team via Slack, create ServiceNow INCWithin 15 min
P3Post to #netclaw-alerts, assign to queueWithin 4 hours
P4Log only, queue for next business dayNext business day

When unsure about severity or approach: escalate. Say "I'd recommend verifying this with a human engineer before proceeding."

Proactive Behaviors

  • When you see a device with CPU > 90%, flag it even if not asked
  • When a CVE matches a running software version, mention it in context
  • When NetBox data doesn't match device state, note the discrepancy
  • When an OSPF adjacency is in a non-FULL state, investigate automatically
  • When a BGP peer is in IDLE/ACTIVE, check both sides

Session Hygiene

  • Start every session by listing devices: pyats_list_devices
  • End every session with GAIT log
  • Don't carry assumptions between sessions — verify current state
  • If a session is interrupted, the next session should check for incomplete changes

Active Technologies

  • Spec153 source development: Python 3.10+ stdlib setup engine and canonical catalog; TypeScript 5.9/VS Code >=1.109 bootstrap, wizard and private assistant broker. New empty homes, separate per-claw credentials, native exact-plan review, unknown-outcome pause and structural report redaction. End-user deployment skill: installer/assistant-skills/netclaw-install/SKILL.md; usage and open qualification: docs/GETTING-STARTED-VSCODE.md and specs/153-ide-installer/verification.md. The owner authorized merge, publication, completed-branch cleanup and return to main on 2026-10-24 after three additional actual Mac installer passes. This supersedes the earlier source-only release hold. Existing owner runtimes remain untouched; Windows/WSL qualification limits remain explicit. See specs/153-ide-installer/release-verification.md.
  • Spec 149 implementation: selected-runtime NCFED adapters and additive harness cards; isolated Python 3.12 federation/MCP, pinned Python 3.14 Hermes companion, separate private federation/HUD state and SQLite execution evidence. Mac controlled-provider acceptance covers all four internal runtime pairs, Hermes/Hermes and Hermes/OpenClaw external execution and authenticated mobile Ask Border. See spec149 verification; Linux/WSL production control and physical mobile qualification remain pending.
  • TypeScript 5.9 / VS Code ^1.102; browser JavaScript; HUD Node >=24.19 <25 or >=26.1; existing isolated Python helpers. + Existing Three.js 0.185.1, esbuild, Express, yaml CST and js-yaml. No adaptivecards runtime or new framework. (151-extension-experience)
  • Existing testbed YAML and selected runtime .env/config; private backups and setup evidence. No draft secrets persisted in webview state. (151-extension-experience)
  • TypeScript 5.9; existing Node 24.19+ ESM helpers; VS Code >=1.102; browser JavaScript. + Existing VS Code APIs, Node standard library, yaml/esbuild and existing HUD runtime/CLI helpers. No new framework or broad shell endpoint. (152-workspace-claw-controls)
  • Existing private installation identity; per-installation management bindings/evidence/backups with 0700 directories/0600 files. Editor stores only non-content binding metadata locally. Draft text remains in memory. (152-workspace-claw-controls)
  • TypeScript 5.9 for the extension (existing esbuild build). Python 3.10+ standard library for the setup engine, plan validator and catalog tooling, consistent with installer-preflight.py and installer-readiness.py. Bash for the existing step functions. JSON Schema 2020-12 for the Claw Plan and the catalog. + Existing VS Code APIs (webviews, walkthroughs, terminals, native input and modal dialogs, MCP server definition provider) and the existing MCP SDK in dist/assistant.mjs. Existing installer libraries, readiness, preflight, component-launch.py, in2n-* helpers and private-file writers. No new Python runtime dependencies for the engine; no new UI framework. (153-ide-installer)
  • Claw Plans live where the user chooses (deployment workspace or a private setup folder) and contain no secrets. Setup-run journals remain authoritative in a private state root outside source/runtime homes; after creation, <runtime-home>/netclaw-hud/setup/ contains only an owned reference and sanitized evidence (R7). All of these use 0700 directories and 0600 files. Secrets go only to runtime or member environment files through existing writers. Editor state holds only non-secret wizard answers and run references. (153-ide-installer)
  • Python 3.10+ standard library for setup engine/catalog/validation; TypeScript 5.9 and existing browser JavaScript for VS Code. + Existing VS Code APIs, esbuild, JSON Schema subset validator, Python unittest, Node test runner and existing runtime CLIs. No new UI framework or general shell endpoint. (155-border-channel-setup)
  • Non-secret schema-1 plans; private per-installation environment/configuration; existing external setup-run journal and owned artifact records. No bulk environment import or credential values in frontend state. (155-border-channel-setup)
  • Python 3.10+ standard library; existing TypeScript 5.9 extension. + Existing JSON, dotenv writer, installer and native config contracts. (156-openclaw-env-defaults)
  • Source JSON defaults; existing private runtime configuration and environment. (156-openclaw-env-defaults)
  • Existing JavaScript ESM/Node24.19.0. + Express and existing private atomic writer. (157-hermes-canvas-retry)
  • Existing owner-private JSON bindings,30-day expiry,8MiB read/write bound. (157-hermes-canvas-retry)
  • JavaScript ESM/React18; Node24.19 tests. + Existing React, Web Crypto SHA-256, esbuild/JSDOM test harness. (158-hermes-chat-recovery)
  • Existing installation-scoped sessionStorage pending record and sanitized chat/archive. (158-hermes-chat-recovery)
  • Python 3.10+ bridge and JavaScript ESM/React; HUD Node24.19+ + Existing SQLite, httpx/FastMCP, Express, React; no new dependency (159-hermes-uncertain-capacity)
  • Existing private HUD ledger and owner bindings; additive release table (159-hermes-uncertain-capacity)
  • Python 3.10+ standard library for installer/normalization; isolated Python 3.12 pyATS/Genie/Unicon and MCP 2.3.0; TypeScript 5.9 and existing VS Code APIs. (160-dynamic-pyats-testbeds)
  • Secret-free Claw Plans and private per-claw environment; process-local dynamic candidates, private per-process task evidence, no dynamic-to-static fallback. See specs/160-dynamic-pyats-testbeds/verification.md. (160-dynamic-pyats-testbeds)
  • Spec162 planned: Python 3.12 for isolated MCP execution; Python 3.10+ compatible installer utilities; existing TypeScript 5.9 and Node runtime policy for VS Code/HUD. + Existing pinned FastMCP 4.0.11 and official MCP SDK 2.3.0; stdlib SQLite/hashlib/path utilities; existing UI stack. Official Terminal 0.5.0 is operator-installed, not vendored or patched. (162-netconverterai)
  • Spec162 planned: private installation-owned SQLite request/attempt ledger and bounded artifact directories (0700 directories, 0600 private metadata; frozen inputs not writable through exposed tools). Existing API-owned Local/Lab records remain authoritative for candidate application. (162-netconverterai)
  • Python 3.12 isolated protocol/n2n runtime; Python 3.10+ installer; existing Bash and HUD JavaScript. + Existing asyncio/federation stack; pyOpenSSL 26.4.0, cryptography 50.0.2, service-identity 26.1.0. Preserve separate certificate-helper environment. No new MCP server or web framework. (168-ngrok-transport-contract)
  • Private installation-owned schema1/ngrok and schema2/direct-or-ngrok registries, atomic revision updates and explicit prior-revision recovery; owned daemon/ngrok process records. Direct GCP transport-only pilot and owner-selected open read-only lab HUD; users/SSO/RBAC remain R15–R17. No tokens in registry/bundles. (168-ngrok-transport-contract)
  • Spec171 planned: isolated Python 3.12; existing Node 24.19/React 18 HUD, cryptography 50/P-256, SQLite, asyncio and MCP. (171-execution-parity)
  • Spec171 planned: additive federation.db tables, private artifacts and owner authority, existing GAIT audit. (171-execution-parity)
  • Python3.12 federation/native tools; Python3.10+ role/IaC helpers; existing JavaScript ESM/React18 and qualified Node24.19. + Existing strict TLS/federation, FastMCP/MCP, SQLite, Express/React/Vite build; OpenTofu1.13.2 Google8.9.0, Caddy2.11.7. No new MCP server or UI framework. (172-gcp-hybrid-border)
  • Selected private homes/SQLite authority and ledger; one-device frozen profiles; external0700 private cloud state/evidence; bounded installation-scoped browser reference storage. (172-gcp-hybrid-border)
  • Python3.12 federation/MCP; Python3.10+ host helpers; existing Node24.19/React18 HUD. + Existing FastMCP/MCP, requests, P256, SQLite, Docker, OpenClaw2026.7.1-2, strict iN2N, OpenTofu1.13.2/Google8.9.0, Caddy2.11.7. Existing NetBox/subnet MCPs; no new server/framework. (173-gcp-cloud-member)
  • Private SQLite/GAIT and immutable profiles; external0700 state/evidence and0600 secrets; bounded browser identifiers only. (173-gcp-cloud-member)
  • existing TypeScript5.9/Node24.19; Dart/Flutter current project toolchain; Python3.12 federation. + VS Code APIs, Node HTTPS, existing Flutter HTTP and secure storage, existing Express owner service. (174-cloud-client-integration)
  • non-secret profile/reference state, native secure sessions, memory-only CSRF/tickets. (174-cloud-client-integration)
  • Python3.12 federation; existing Node24.19/React18 HUD; TypeScript5.9 VS Code; Dart3.12.2/Flutter3.44.8 and Swift under installed Xcode27.2. + existing SQLite, qrcode, WebSocket/TLS/P256, Express, VS Code APIs, Flutter secure storage/mobile_scanner. No new server/framework or hosted QR service. (175-cloud-mobile-enrollment)
  • additive private enrollment issuance records and existing private federation authority; device secure storage for keys/pending identity/references; QR matrix only in transient UI memory. (175-cloud-mobile-enrollment)
  • Spec176 planned: Python3.12 federation/runtime; existing Node/React/TypeScript/Flutter clients; existing MCP, SQLite, Docker, OpenClaw and DefenseClaw with pinned Ollama0.40.11/Qwen3:0.6b CPU profiles. (176-ollama-members)
  • Spec176 planned: additive private parent/profile records, existing child authority, per-request claims and member/runner-wide unknown-provider latch; bounded client references. (176-ollama-members)
  • Python3.12 federation; Python3.10+ host helpers; existing Node24.19/React18 HUD; TypeScript5.9 VS Code; existing Flutter/Dart. + Existing asyncio/SQLite/cryptography/pyOpenSSL, FastMCP/MCP, pinned OpenClaw gateway and subnet MCP, Express/React, VS Code APIs, Flutter secure storage/scanner, OpenTofu1.13.2/Google8.9.0 and Caddy2.11.7. Reuse verified public artifact closures. (177-external-en2n-federation)
  • Separate private homes, federation DBs, keys, gateway state, existing grants/tasks/chat/GAIT. Bounded non-secret client provenance. Plans/evidence outside source in0700/0600 storage. (177-external-en2n-federation)
  • Python3.12 federation; existing Node24.19/React18 HUD; TypeScript5.9 VS Code; Dart3.12.2/Flutter3.44.8 Mobile. + Existing asyncio/SQLite, cryptography/pyOpenSSL, FastMCP/MCP, Express/React, VS Code APIs, Flutter secure storage/WebSocket. (178-recovery-gate)
  • Existing private identity/ledger/reference stores; additive bounded identifiers and monotone send markers; private evidence0700/0600. (178-recovery-gate)
  • Python3.12 federation/runtime;Python3.10+host helpers;existing Node24.19/React18,TypeScript5.9,Flutter/Dart. + Existing NCFED/FastMCP/MCP,asyncio,httpx,cryptography/pyOpenSSL,SQLite,Docker;external pinned etcd3.6.16. Existing OpenClaw/DefenseClaw/Ollama qualification closures. (179-border-ha-authority)
  • Private etcd bounded metadata plus member SQLite claims and immutable signed artifacts;0700directories/0600private files. No credentials in frontend state. (179-border-ha-authority)
  • Spec 148 implementation: JavaScript ESM/Express/React HUD; HUD Node >=24.19 <25 or >=26.1; isolated bridge/tool Python 3.12 and companion Python 3.14; official MCP Node client 2.3.1 and isolated FastMCP 4.0.11/MCP Python SDK 2.3.0. General OpenClaw installer Node policy remains >=24.16 <25 or >=26.1.
  • Qualified Hermes baseline: v0.21.6, source 818c13be1dc4fd28987e1e881a9408224afd4535. Private MCP bridge/protected companion, schema-2 owned HUD bindings, SQLite admission/evidence records and installation-scoped browser storage. Mac and Ubuntu 26.04 WSL live/browser qualification is recorded in specs/148-hermes-hud-integration/closure.md; broader tool and platform qualification is not implied.

Recent Changes

  • 148-hermes-hud-integration: all 58 tasks complete for the owner-approved Mac/WSL scope; source version 1.7.0. Live Hermes/provider, browser, process-fault, preservation and existing OpenClaw evidence is in specs/148-hermes-hud-integration/closure.md. Ubuntu 24.04 and native Windows OpenClaw remain explicitly unverified. Hermes n2n/iN2N federation is separate spec 149.

Spec151 implementation notes

The extension now has local source-development cards/forms, testbed YAML/JSON/CSV management, scoped MCP configuration/setup and native-bound security controls. See specs/151-extension-experience/verification.md for actual evidence and platform limits. Distribution remains held by the owner. DefenseClaw's current connector-scoped rules and revision-aware config writer differ from historical workspace/skills/defenseclaw-ops/SKILL.md source-scoping examples: capability-probe the installed CLI; do not infer an MCP-only policy scope from --source. OpenShell lacks an atomic revision precondition; coordinate external writers and never claim universal blocking from one removed allow rule.

Spec151 repository merge was explicitly authorized on2026-10-11 after local completion. Commit/push/PR/merge may proceed; packaging, deployment and release/Marketplace publication remain held.

Spec152 production release was explicitly authorized on 2026-10-16 after the WSL inspection. Fetch/commit/PR/merge, completed-branch cleanup and production publication may proceed with verification. Historical release holds for Spec152 are superseded.

Spec155 production publication was explicitly authorized on 2026-10-25 after Mac reconciliation. Commit/PR/merge, versioned source and Marketplace publication, completed-branch cleanup and return to main may proceed. This supersedes earlier Spec155 source-only holds. The release retains T048 and overall 49/50 qualification; six WSL service-manager failures, genuine remote-editor acceptance and documented fixture limits remain visible. Existing installations and provider/model/device operations remain outside release scope. See specs/155-border-channel-setup/release-verification.md.

Spec160 release authorization (2026-10-30): the owner requested versioned source and touched-extension publication, README/community notes, commit/push/PR/merge, completed-branch cleanup and return to main. Earlier Spec160 source-only publication holds are superseded. The two-part modern_website writeup remains staged as drafts; no website deployment or existing-runtime upgrade is requested. Preserve all recorded qualification limits.

Spec162 NetConverter release was explicitly authorized on 2026-11-01: reconcile upstream, publish the versioned source and touched extension, PR/merge and return to main. Source 1.18.0 combines this feature with upstream 1.17.1; extension is 1.5.0. This supersedes the earlier unpublished source milestone hold. The website article remains a staged draft, and the leadership/community messages remain unsent. Existing running lab/Risk installations are not upgraded. See specs/162-netconverterai/release-verification.md for exact scope and qualification limits.

Spec168/R01 closure (2026-11-01)

The owner fully delegated the isolated R01 cloud/DNS/HTTPS activation, relocation, full open lab HUD and scoped Anthropic Sonnet demonstration. R01 is40/40 for that amended scope; stop here. R02/R03, source commit/push/release, existing-runtime upgrade and device operations are not authorized. Live lab URL and correlated cloud-to-on-prem evidence: specs/168-ngrok-transport-contract/verification.md and onprem-answer-evidence.md. Cloud Border and fresh Mac subnet member use Sonnet4.6; the member is not the existing pyATS installation. Actual browser Chat, RAG indexing/retrieval, cited Canvas handoff/answer, unchanged-key relocation, DNS/trusted HTTPS and no-drift checks passed. Optional ngrok stays supported; live ngrok and the explicitly untested platforms remain unqualified. The full public lab exception does not implement users/SSO/RBAC (R15–R17). Private operator access stays IAP; no raw admin/device-write API is published. Preserve the authoritative private pilot.tfstate and active homes; never replay old activation plans or initial secret payloads. VM stops December1; disk/IP costs remain until separately reviewed teardown. GAIT and private receipts retain the full trail.

Roadmap source integration and release hold — 2026-11-02

The owner authorized committing, pushing and merging the completed R01 source into main, with a short maintainer summary. This supersedes R01's earlier source-merge hold only. The issue310 roadmap work remains unreleased until the entire roadmap is complete, including user identities, SSO and RBAC (R15–R17) and closure (R20). Do not bump a version, create a release/tag, publish a package/Marketplace build, publish a blog/changelog/announcement, or promote general cloud deployment for this work. R01 transport proof is not authorization to deploy a cloud Border without user controls. Keep technical specs/evidence for maintainers; rollout copy is deferred under specs/168-ngrok-transport-contract/deferred-docs/ and absent from the public README, setup wizard and general installation guides. Existing owner pilot authorization is a bounded exception, not a user deployment recipe. R01 integrated on upstream 1.18.1 without a version increment. Subsequent upstream fixes #317/spec169 and #318/spec170 advance source to 1.18.3; extension 1.5.0 is unchanged. These maintenance fixes do not release or qualify the cloud roadmap. Next roadmap brief is R02 execution parity, expected next available spec 171 (recheck at kickoff; 169 and 170 are occupied by those upstream fixes); it is not started by this integration request. No R02 cloud/device operation is authorized here. Keep the original R01 directory name despite the parallel upstream168-openclaw-2026-10-compat feature; cite R01/full path to distinguish them.

R02–R05 work authorization — 2026-11-02

The owner subsequently authorized driving R02–R05 to completion with a coordinator and three parallel agents, using separate worktrees. This supersedes the earlier R01-only stop and R02 source-work hold. Human roadmap acceptance gates and exact cloud-resource/spend reviews remain. Total GCP lab spend must not exceed $100/month; use CAD100 conservatively, including R01, pending any owner correction. Budget alerts are not hard spending caps. Optional bounded implementation/test/review passes are capped at five total across R02–R05. The entire-roadmap release, version and public rollout hold remains in force. Spec171 implements R02; R03 cannot begin before its human evidence gate.

R02/Spec171 technical acceptance — 2026-11-02

Source8c178ff8 passed both actual protected OpenClaw/Hermes Sonnet reads through fresh Mac/Docker members: one R1show-version, zero setup commands, signed results, durable ACK, attributable GAIT and no-repeat restart recovery. Independent review passes. All36 tasks are complete; the owner accepted T036 on 2026-11-02 and authorized R03 planning. Exact cloud-resource/spend review and later human gates remain. See specs/171-execution-parity/verification.md, parity-matrix.md and resume.md. Two of the five bounded implementation passes were used. The earlier unknown SSH operation and authentic reads with an implicit Unicon setup probe remain retained with their limitations, never replayed. The corrected native policy is narrowly pinned to IOS-XE/IOL and the qualified Mac dependencies. No arbitrary provider, cloud, other-plugin, write or multi-user qualification is implied. Source1.18.3 and extension1.5.0 unchanged; entire-roadmap release hold remains. No R03 resource or existing R01 runtime was changed.

R03/Spec172 source checkpoint — 2026-11-02

The owner requested driving R03 to100%. Source is integrated on 172-gcp-hybrid-border: immutable four-device selection, production owner HUD, strict control-only Border role/posture, one-use protected enrollment and isolated GCP IaC/host controls. Combined protected/posture464 tests and fullHUD453 tests pass; actual LinuxAMD64 dependency extraction is verified offline. Four lab endpoint/host-key diagnostic proofs are private, with explicit console versus historical SSH-pin provenance. No actual R03 cloud apply/activation/provider read, DNS change or R01 runtime change has occurred. R03 is not100% or human accepted. Current billing/total CAD100 gate, exact cloud/runtime delegation, actual GCP controls/HTTPS/four protected reads/recovery and reviewed teardown remain open. The optional exact R01 stop in Spec172 cost-bound-review.md is awaiting owner choice; do not execute it without delegation. Source1.18.3/extension1.5.0 and roadmap release hold remain unchanged. Pass3 source/offline preparation is complete; actual execution remains gated. No automatic Ralph hook. Exact foundation apply delegation was requested and is pending; no elapsed time supplies approval. See Spec172 verification.md/resume.md and private roadmap-r03-20261102 evidence.

R03 activation delegation — 2026-11-03

The owner explicitly said to do what is needed, push forward using the stated GCP credit, stop treating money as the blocker, and clean up owned resources so charges do not continue. This supersedes the Nov2 billing/approval holds for the reviewed isolated R03 foundation/runtime/DNS/four-read/verification/cleanup scope. Do not repeat those permission requests. Preserve exact plans, bounded safeguards, unknown outcomes and audit evidence; the credit is not a spend target. Cleanup follows evidence export and includes billable residues. R01/R02 work and unrelated runtimes stay preserved. Release/version/public rollout holds and R04 human gate remain. See specs/172-gcp-hybrid-border/activation-delegation.md.

R03 technical qualification and retirement — 2026-11-03

R03 implementation, actual GCP/on-prem qualification and reviewed cleanup are complete;35/36 tasks closed. T036 memory/GAIT is recorded but human acceptance remains pending before R04. Four browser-approved R1/R2/SW1/SW2 show version reads returned IOS17.18.2 with one command/two OpenClaw-Sonnet provider calls each, signed results, durable ACKs and unchanged effects after loss/restart recovery. 62 browser/auth checks and actual cloud confinement/IAM/network checks passed. See Spec172 verification.md for narrow platform/provider scope and retained failures. All13 R03 cloud resources, owned DNS and temporary OS Login key are removed; actual inventories found zero owned billable residues. Local workers and owned containers/networks are retired; private homes/current-state archive remain. R01 state/DNS and running pilot are unchanged, with existing costs retained. Never replay spent R03 plans, enrollment or first-create state; use the private current-state preservation record only as evidence, not an activation payload. Source1.18.3/extension1.5.0 and the entire-roadmap release hold remain unchanged.

R02/R03 source integration and R03 acceptance — 2026-11-03

After receiving the completed R03 technical/cleanup handoff, the owner requested commit, PR, quiet merge and return to main, plus a small maintainer blurb. This accepts the completed R03 scope and closes T036 (36/36); no new R04/cloud/device work is implied. Commit/push/PR/merge of completed R02/R03 source is authorized. Reconcile upstream #321 failure reporting and retain its source1.18.4 version; extension1.5.0 remains unchanged, with no additional roadmap bump or publication. The complete-roadmap release hold through R20 is still in force. Keep historical 171-execution-parity (R02) and upstream 171-skill-failure-reporting distinct by full path; never rename historical specs to hide the collision. R03 live evidence remains tied to its frozen tested source, not a claim that merged source was redeployed. See specs/172-gcp-hybrid-border/source-integration.md for verification and PR outcome.

R04 autonomous execution — 2026-11-03

Owner requested R04 to100%, supplied the existing public NetBox demo source and private credential, required actual NetClaw member/NetBox skills/MCP, then delegated autonomous completion. Use fresh isolated cloud Border/member and separate subnet, read-only device/prefix/subnet workloads, finite bounds, real evidence and full owned cleanup. Existing source guard pins select Sonnet4.6 instead of historical Haiku default. Do not repeat delegated permissions. Preserve R01, retired R03 archives and source1.18.4/extension1.5.0; no release/public rollout or R05. Pass4 is R04; see specs/173-gcp-cloud-member/activation-delegation.md.

R04a and continuing client integration — 2026-11-03

The owner inserted R04a between accepted R04 and R05 to catch up VS Code cloud Border connectivity and affected Mobile flows. See docs/roadmap/issue-310/briefs/R04a-cloud-border-client-integration.md. Every subsequent stage must separately assess, implement where affected and verify VS Code and Mobile integration under roadmap G12/G13; record evidence for no-impact decisions instead of silently deferring client work to R20. R04a is planned, not started by this documentation amendment. Preserve current R04 qualification/cleanup, human gates, existing milestone IDs and release holds.

R04 technical qualification and retirement — 2026-11-03

R04/Spec173 implementation, actual cloud-member qualification and owned cleanup are complete. The existing NetClaw NetBox skill/MCP read the observed demo device and prefix; the existing subnet tool calculated192.0.2.0/26. Three separate browser approvals produced three native MCP calls, six guarded Sonnet4.6 provider calls, signed results and durable ACKs. Actual member loss/restart preserved the same authorities/results/effect counts without replay. Two earlier attempts remain unknown with zero provider/MCP calls and their original witnesses. Driver and archive failures and all narrow qualification limits remain in verification.md.

All24 owned R04 cloud resources, DNS and temporary OS Login access are removed. Actual12-category inventory found zero owned residues; R01's checked resources, state, DNS, HTTPS and shared budgets remain preserved, with R01 costs retained. Private current-state archives and the1423-file frozen export are local under ~/.local/share/netclaw/roadmap-r04-20261103/. They contain private authority and are evidence-only, never activation payloads. Do not replay spent R04 plans, enrollment or attempts, or reconnect to its retired hostname/IPs.

Human acceptance remains required before R04a, then R05. Pass4 of five bounded R02–R05 implementation passes is complete; no automatic loop. Source1.18.4 and extension1.5.0 are unchanged, with the entire-roadmap release/public rollout hold through R20 intact. No R04a, later roadmap, existing-runtime upgrade, release or external announcement was performed. See specs/173-gcp-cloud-member/verification.md and resume.md. This does not self-record the owner's acceptance.

R04 acceptance and source integration — 2026-11-03

After the completed R04 qualification/retirement handoff, the owner explicitly requested full commit, PR, merge, completed-branch deletion, return to main and a short R04/R04a/R05 writeup. This accepts the completed R04 scope (33/33) and supersedes its pending acceptance/source-integration gate. Preserve all frozen live evidence and original qualification limits; merged source is not a claim of redeployment. Include the owner's R04a and continuing VS Code/Mobile roadmap amendment in this integration. R04a is next, then R05; neither starts here. Source1.18.4/extension1.5.0 remain unchanged. No version, release, Marketplace, blog or public rollout authorization; the entire-roadmap hold through R20 remains. See specs/173-gcp-cloud-member/source-integration.md.

R04a client integration authorization — 2026-11-03

The owner subsequently instructed driving R04a efficiently to completion. This starts Spec174 on 174-cloud-client-integration; R04 is already accepted and merged. The coordinator and separate VS Code/Mobile lanes implement explicit cloud-owner clients using the existing protected HTTPS authority, exact review and no-repeat recovery. Fresh finite qualification and owned cleanup remain scoped to the R04a client evidence; never replay retired R04 state, identities, enrollment or workloads. Preserve R01 and existing owner clients/runtimes. Four optional roadmap passes were used; this is pass5 with no automatic extra improvement loop. Required verification/fixes are part of this pass. Actual editor/native Mobile evidence and human review precede R05; inventory, source tests and simulator execution cannot imply physical-device qualification. Source1.18.4/extension1.5.0/Mobile1.0.3+6 and the R20 publication hold remain.

R04a technical qualification and retirement — 2026-11-03

Spec174 technical work is complete:16/17 tasks, with human evidence acceptance still required before R05. Actual Mac VS Code1.109.0 and native iOS26.5 simulator cloud-owner reads each produced one NetBox MCP call, two guarded Sonnet4.6 calls, a signed result and durable ACK. Actual process restarts recovered the same references/results with GETs only and no repeated effects. Three unapproved original reviews remain owner-denied with no execution. Physical iPhone remains unqualified because installation was refused while locked. The final narrow Mobile definite-unsubmitted recovery fix has20 focused tests and independent review; successful live evidence retains its earlier source binding.

All24 fresh R04a cloud resources, owned DNS and temporary OS Login key are removed; 12 actual residue categories found none. R01's checked resources/state/DNS/HTTPS and shared budgets remain preserved, with its existing costs retained. The315-file private frozen export is evidence only, excludes private identity keys and selected integration credentials, and must never be replayed as activation state. Never reconnect to retired R04a addresses or replay its plans/enrollment/approvals. See specs/174-cloud-client-integration/verification.md and resume.md for exact platform/source, preservation and failure limits. Pass5 is complete, with no extra optional loop. No R05, version change, release, public rollout or existing-runtime upgrade was performed. This does not record the owner's acceptance.

R04a owner-delivery correction — 2026-11-04

The owner rejected the disposable-only completion as failing the actual objective: his installed VS Code must connect to a persistent usable GCP Border. He explicitly authorized rebuilding the GCP NetClaw Border image if necessary and designated this as non-live development/build-time infrastructure. Reopen R04a. This supersedes the earlier no-existing-runtime-upgrade restriction for the scoped GCP pilot and necessary private local client installation; it does not authorize public release, Marketplace publication, unrelated runtimes, device writes or R05. Retain concrete baseline/rollback, review exact resource changes and preserve the existing trial expiry/spend bounds. Leave the intended owner deployment and client profile usable; retire only disposable test resources. Prior isolated evidence remains historical, not owner-delivery acceptance. No new optional improvement loop is started: this is required correction of the unfinished R04a objective.

R04a connection-flow and authentication correction — 2026-11-04

The owner explicitly rejected the premature authentication requirement: no login, one-use tickets, owner sessions or another auth scheme in this current development pilot. Authentication belongs to the later roadmap steps and will use a different design. The exact requested VS Code flow is Add Connection → Local / Cloud → Cloud → Name → URL → Connected. Do not ask for installation ID or an access-mode choice; discover and pin installation identity over verified HTTPS. Selecting the cloud profile must open a useful view directly and must not flicker/reconnect an already-selected profile. This supersedes earlier R04a owner-sign-in assumptions and requirements for this pilot. No hidden auto-login/ticket mechanism may be substituted. Preserve exact read scope/review, no-repeat recovery, private member credentials, TLS/endpoint identity and truthful unauthenticated lab attribution. Existing approval/history must not be relabeled or silently cleared. Keep later identity/SSO/RBAC and public rollout deferred; no general deployment/release is authorized. Owner supplied a replacement NetBox key privately; use it instead of the temporary demo credential created during diagnosis, and never put it in source or output.

R04a retained runtime checkpoint — 2026-11-04

The explicit no-user-auth Border/member migrations and corrected private extension installation are complete. Same Montréal URL/VM, December1stop, original identities and failed403 authority are preserved. Cloud public readiness0e1ef7c3 and member verification7e24664d pass. Owner replacement key is installed only on the member; source1.18.4/extension1.5.0/Mobile1.0.3+6 unchanged, no public release.

Actual native completion is blocked by the Mac lock screen. The owner has been asked to unlock normally; never request a password or send blind UI input. After unlock, use only extension-host restart (preserve CodePID924/window/terminal), exact Name+URL connection and stable profile selection, then same original5004 recovery under GAITb21011f2. New distinctprefix1 approval requires full root review; signed success/ACK plus GET-only restart/effect comparison remain required before100%. Original apply and completion claims are spent; never rerun them or restore oldDB. See specs/174-cloud-client-integration/resume.md for exact remaining sequence. Existing Mobile QR needs proposedR04b HUD/WSS/protectedoperations; preserve scanner and devicekeys, no replacement manualwizard or userauth. Human acceptance/R05 and roadmap release holds remain. This checkpoint is not a completion or acceptance claim.

R04a actual operating-interface correction — November 4

The owner explicitly rejected the JSON-only connection landing and narrow read-operation client: make the normal NetClaw VS Code interface work against the cloud Border after Name + URL, with no user authentication. This supersedes R04a's earlier operations-only UI acceptance assumptions. Do not count a connected badge, diagnostic document, menu presence or isolated read as completion, and do not build another operations-only substitute. Stop redundant verification of that rejected target. Reuse normal Overview/Chat/Canvas/RAG/estate/integration views and the existing full HUD/runtime handlers through verified HTTPS remote capabilities. Local extension-host filesystem/install/lifecycle actions remain local. Preserve current b164 installation, protected authority and failed403 history while composing any separate conversation/runtime services; do not silently restore old f03c authority or legacy device execution. Needed owner-pilot development runtime updates remain delegated; no additional user sign-in, release/version/public rollout or R05. Native Name+URL/no-login/stableclick checks passed before this correction; initial extension activation passed with CodePID924 preserved. Original5004 reconciliation was NOT submitted. Actual full operating-interface qualification remains open.

R04a live workspace checkpoint — November4, final recovery

The normal cloud workspace is now deployed and active in the owner's original VS Code window. Private VSIX3fa30a37 (clientcebf8109/root7856e856) preserves profiles, settings and other extensions. Actual normal Overview, Chat, RAG retrieval/context review and cited Canvas passed; the sole342-byte upload used production HudClient after native file-picker automation failed before submission. Five public docs and gateway logs were repaired; all advertised read panels and real Sonnet usage passed. No user authentication was introduced. Current cloud gateway56278/HUD56057/edge55882 and protected federation51900/Caddy50879 remain running; sourcee68+7856delta, same b164 identity, authority and December1stop. No new cloud resources or publication.

Original5004 is signedFAILED/native_netbox_http_403 with durableACK and unchanged 1provider/1MCP/1GET403. Authoritative settlement is17:36:49UTC, initiating caller unestablished; earlier summaries of an unresolved operation were stale. Current native Recover GET repeated no action. Fresh prefix9c380fdc was approved once and succeeded with2guardedprovider/inference calls,1nativeNetBoxMCP/1HTTP200GET, signed resultbae4366c and durableACK; both member and Border pre-recovery collectors pass.

Final correctly targeted extension-host restart39253/39255→57203/57208 and existing profile reconnect succeeded; mainCode924, both references and completed Chat/Canvas files are unchanged. macOS locked before final operation selection. The owner has been asked to unlock normally and leave the original netclaw window visible. After unlock finish ONLY same-prefix GET recovery and normal Overview, then run the existing recovered-phase collectors. No more restart, preparation, approval, reconciliation, upload, search or provider prompt is needed. Do not claim the final native lookup passed until it does. See specs/174-cloud-client-integration/resume.md. Human acceptance beforeR05, proposed Mobile QR R04b and R20 release holds remain.

R04a final owner verification — November4

The owner authorized the last check after unlocking. Actual native post-restart status at20:03:24UTC displayed the same prefix9c380fdc completed, signed result verified and delivery acknowledged; normal connected GCP Overview is left visible (20:04:31UTC). MainCode924/hosts57203/57208 remain unchanged; all five persisted NetClaw files, both references/decision markers and completed Chat/Canvas match their pre-restart hashes. Native final receipt1e9c306c retains the exact-click attribution limit: the picker was selected between snapshots; the source path is GET but no raw HTTP trace is claimed. No extra lookup was forced.

Root's member recovered-phase collector verified identical approval/result/ACK, unchanged original403 evidence and new-prefix2provider/2inference/1MCP/1HTTP200GET. Border comparisonf52c7fc8 verifies every retained field unchanged except observation timestamp, including both authorities/results/ACKs, GAIT, counters and source pins. All21 technical R04a tasks are complete; T017 remains the separate human acceptance gate beforeR05. Do not resume old pending-lookup instructions or run another workload/restart to improve evidence. Keep the intended pilot/member running within its December1scope. Mobile QR is proposedR04b; no R05, version bump, publication or general rollout occurred. Final handoff is specs/174-cloud-client-integration/resume.md.

R04a acceptance and production extension update — November4

After the completed native verification handoff, the owner requested commit, PR, quiet merge, return to main and production VS Code packaging/publication. This accepts R04a (22/22 including T017) and supersedes the earlier extension version and Marketplace hold for this scoped update. Preserve upstream source1.18.5 from PR325 and both distinct Spec174 directories. Extension1.6.0 targets the verified Marketplace1.5.0 baseline; publish only the checked archive with recorded source and digest. Verify PR checks, merge and public artifact availability.

The broader source release/tag, blog, external announcement and general cloud deployment holds throughR20 remain. Do not redeploy the retained pilot or restart the owner's editor, repeat provider/device operations, or start R04b/R05 during this integration/publication task. R04b cloud HUD Mobile QR enrollment is the recommended next scoped feature, followed by R05. No new user-authentication scheme belongs in either client correction; users/SSO/RBAC stay R15–R17. See specs/174-cloud-client-integration/release-verification.md for actual release evidence.

R04b kickoff and client scope — 2026-11-04

Owner said “04b time”, then explicitly required easy QR generation directly in local and GCP/cloud HUD, the same enrollment generation in VS Code Mobile, and actual simulated/virtual iPhone verification. Spec175 covers the existing scanner, device keys, scoped one-use QR, trusted WSS, protected operations and recovery. Use coordinator and separate backend/surfaces/Mobile worktrees. Keep the owner's usable pilot, existing authorities and December1 lifetime; clean only disposable test resources. No new user login/SSO, later roadmap stage or publication is implied. Native simulator image decode/key fixtures must stay distinct from physical camera and SecureEnclave qualification. See specs/175-cloud-mobile-enrollment/.

R04b completion boundary — 2026-11-05

The owner explicitly requested: “get this to 100% and full stop before proceeding to R05.” Complete Spec175 implementation, actual local/cloud HUD and VS Code QR enrollment, native simulator operation/recovery verification, evidence and cleanup. Then stop. Do not start R05 or another spec. This is execution direction for the current stage; it does not pre-certify unobserved tests or authorize publication.

R04b native qualification checkpoint — 2026-11-05

All four actual local/cloud HUD and VS Code QR paths passed on the owned iPhone simulator, including retained-key reconnect and production navigation. Native binding/scope refusals and cold/background recovery passed. The private1.6.0 extension is installed, with settings/profiles/other extensions preserved; no Marketplace publication. The exact installed artifact passed actual Overview and Mobile checks with a cloned existing cloud profile. Original editor/terminal remained untouched; macOS accessibility was unavailable.

The reviewed pilot/member update is settled and preserved authority/history. One native approved prefix1 read returned signedFAILED native_netbox_http_403, with durable ACK and exactly1 provider/1 MCP/1 HTTP request. Recovery repeated no effect. The unchanged Nov4 demo token likely expired across the nightly reset; exact403 cause is unproven. A current authorized read-only token is required for successful-read acceptance. An async request for its private local file path is pending. Do not create external tokens or replay the failed operation. Use a separate reviewed reference and new evidence directory after a valid credential is supplied. Spec175 is not100% or human accepted; see its resume.md.

Owned disposable services/editors/simulator are stopped. The failed attempt's3 containers and1 network were removed after verified export; original101 containers, 50 networks, all operation records and source evidence remain. The usable pilot and member retain December1scope. All deployment/cleanup claims are spent; no DB restore or replay. Full stop beforeR05 remains in force.

R04b acceptance and scoped release authorization — 2026-11-05

The owner supplied a replacement private read-only credential and then explicitly requested packaging, commit/PR/merge, a source patch release, VS Code publication, changed Mobile app release, release notes, a safe QR screenshot, a short maintainer note, a draft in sibling modern_website, and return to main. This accepts the completed R04b scope and supersedes prior publication/version holds for Mobile QR enrollment only. Source1.18.6, extension1.6.1 and Mobile1.0.4+7 are selected; actual publication outcomes belong in Spec175 release-verification.md. Website deployment and external announcements are not requested. Broader cloud rollout remains held.

The separately reviewed new native operation9902abf3 succeeded with a signed NetBox prefix10.112.0.0/15 result and durable ACK. Cold launch and actual background/ resume preserved the same key/reference/result with no repeated effects: exactly 2 provider calls,2 inferences,1 native MCP call,1 HTTP request and1 attempt. The three historical operations, including the earlier403, remain unchanged. Private success-verification-20261105 evidence is separate from the failed-operation trail.

Root independently reviewed cleanup packet2bb60849 after a verified411-file export; only the new operation's3stopped containers and1empty internal network were removed. Baseline101containers/50networks, all4operation rows and source evidence are preserved. The owned simulator is shut down with profiles retained. The usable pilot/member keep their existing December1scope. Full stop before R05 remains explicit; neither release nor acceptance starts another spec.

R05 / Spec176 SDD kickoff — 2026-11-05

The owner explicitly started R05 after accepted R04a/R04b, requesting Spec176, heavy reuse of Thermidor and then the branch/spec artifacts. This supersedes the prior stop-before-R05 for this stage. Branch 176-ollama-members contains the specification, evidence-derived plan/contracts and33 unchecked implementation/ acceptance tasks. No R05 runtime/cloud/model/client execution occurred in preparation. Use the final Thermidor004 profile and tests, pinned in thermidor-reference.json; never import dirty HA work, Tailscale or expired trial/runtime state. Native subnet source needs bounded IPv4/IPv6 qualification; provider uncertainty requires a member/runner-wide latch as well as per-request claims. Preserve existing pilot, December1 lifetime, source1.18.6/extension1.6.1/Mobile1.0.4+7 and publication holds. Actual runtime changes retain exact scoped review; human R05 acceptance precedes R06/R07. No new optional loop. See specs/176-ollama-members/.

R05 implementation authorization — 2026-11-05

Owner said to go ahead with the prepared Spec176, emphasizing proven/tested Thermidor reuse. Implementation is started; use the existing coordinator and separate-worktree delegation for R02–R05. No extra optional loop or release is authorized. Exact runtime/resource review, preservation and final human gate remain; do not repeat source-implementation permission requests.

R05 owner clarification (2026-11-05): this is NetClaw with NetClaw members using Ollama over existing NetClaw iN2N, no Tailscale. Thermidor provides proven patterns, tests and lessons only; do not deploy/import the Thermidor product/runtime or introduce its transport, identities, authority or lifecycle. Final model/guard images and source are NetClaw-owned, with official pinned public Ollama/model bytes.

R05 exact 48-hour cloud approval — 2026-11-05

The owner selected the48-hour window (~CAD5.47) for the reviewed separate 2vCPU/8GiB Ollama member,40GiB disk and isolated network. This supersedes the proposed December1 retention for the new R05 member only. Revised saved plan ca32d0d28314d3f90cd5aa0cf0ef5d923524d46e69b5cc3a141c54b2e14dba30 applied once:8creates,0changes,0destroys, stop2026-11-07T19:41:25Z. Preserve existing Border/R01 identities/history/DNS/state. Full owned cleanup after qualification must remove disk/IP residues; STOP alone is not cleanup. No publication/version/R06/R07 authorization is implied. Use fresh R05 state under ~/.local/share/netclaw/roadmap-r05-20261105/cloud-plan-48h/; oldDec1plan is evidence-only and must not be applied. NetClaw Ollama members, noTailscale.

R05 technical qualification and retained delivery — 2026-11-06

Spec176 implements fresh NetClaw Ollama members over existing strict iN2N, with Thermidor004 patterns/tests as the foundation and no Tailscale or Thermidor runtime. Actual local ARM64 IPv4/cloud AMD64 IPv6, combined browser, Mac VS Code and native iPhone simulator flows passed: separate exact approvals, native verified facts, signed results/durable ACKs, positive usage, confinement and finite resources. The integrated protected suite passed776 tests; client suites/builds and actual recovery evidence are recorded in specs/176-ollama-members/verification.md.

Preserve the real unknown outcomes and original Mobile ordinal2 CLAIMED inference. Its actual model stop/reattest/recover/profile-rebind is durable; never clear the latch/database or replay old IDs. Two deliberate cloud pause outcomes remain unknown with zero inference/tool dispatch. Actual shared worker restart and all three clients' GET-only recovery preserved authority/results/claims/call counts. Small-model commentary can be inaccurate; native independently verified facts are authoritative and displayed separately. Physical iPhone, other OS/runtime/ model/GPU/HA/eN2N and live injected native-ACK-loss qualification are not implied.

The existing owner Border, original protected history/bindings/keys and unrelated runtimes remain preserved. Final disposable cleanup removed34 stopped containers and11 empty networks after verified export, in addition to the earlier17/6. Original failed helper attempts and one-use continuation receipts remain retained. The intended two models, canonical qualification peers, member homes and owned networks remain for the approved48-hour delivery. Primary private archive: ~/.local/share/netclaw/roadmap-r05-20261105/handoff-export/final-20261105T222010Z/, manifest d80bc85f4c2603002231b9ce29cbc4d2d848086da156d99d9b076f4774fea4e6; later cleanup/current-state evidence is separately sealed, never rewritten into that primary archive. Use current handoff pointers, not historical startup payloads.

Full owned cloud cleanup is armed for November7: quiesce/export19:36:25UTC, exact ten-resource destruction no earlier19:41:25UTC. Current execution manifest 5b6a39412c4d02f8ab9ad3923af9768eef34eddc62cd281820a13f41b7d201ec binds the recovered cloud profile and verified Docker absence handling. It requires this Mac awake/logged-in/online, Docker and gcloud/IAP authorization. Independent VM STOP remains the fallback; disk/IP costs remain if full deletion does not succeed. Future teardown is not yet completed and is not a hard spending cap. Never replay spent activation/recovery/cleanup claims; inspect actual state after interruption. Existing Border lifetime is unchanged. After this technical handoff, the owner explicitly requested completion, commit/PR/merge and return to main on November6, accepting R05 and closing33/33 tasks. Source integration is authorized; do not repeat permission requests. Source1.18.6, extension1.6.1 and Mobile1.0.4+7 remain unchanged; no release/version/ public rollout or R06 implementation is authorized. Preserve both independent Spec176 directories (ollama-members and upstream convergence-vm-security). Next is R06 external federation, then R07 recovery, R08 Border HA, R09 three-region GCP and R10 Azure single region. Recheck the next free spec number at kickoff. See specs/176-ollama-members/source-integration.md for the clear-session handoff.

R06 / Spec177 kickoff — 2026-11-06

The owner explicitly started R06, requested a new branch and full SDD through 100%, and clarified that existing non-cloud eN2N already works: confirm and fix cloud parity without redesigning federation. Branch177-external-en2n-federation uses three independent NetClaw organizations, full triangle/all six directions. Two cloud Borders use separate resources in the existing GCP project; local C is a third organization. Source-planning default is fresh ordinary full Borders, not changes to existing protected homes. Runtime-profile question remains open; no cloud activation or approval is inferred from silence. Exact resource/runtime plans, owner resource preservation, R05 November7 teardown and release holds remain. Research agents are read-only under Spec Kit planning; no implementation delegation or later roadmap stage is implied. See specs/177-external-en2n-federation/.

R06 source and concrete review checkpoint — 2026-11-06

Spec177 source implementation/preparation has24/35tasks complete. Existing eN2N strict triangle, selected-runtime MCP, original-request recovery, external native source attribution and ordinary cloud HUD/editor/Mobile support are implemented. Full federation/gateway/cloud-control1758pass/5skip; HUD518; editor205/3skip; Flutter497/analyzer; native Linux packages and Caddy validation pass. These are source/offline results, not actual R06 cloud/native qualification.

The full concrete activation packet is pending owner review. Fresh run 20261106-a4fd has a saved16-create foundation and a34-create activation preview; no cloud/DNS/provider mutation or apply attempt occurred. Do not apply the combined preview. After actual scoped approval, continue all delegated steps without repeating permission requests; retain exact plans and no-replay evidence. Expiry Nov8 15:19:35Z, base cloud estimate CAD10.95/48h plus disclosed usage. Preserve R05 Nov7 19:36:25Z/19:41:25Z retirement and original Border December1. Source1.18.6/extension1.6.1/Mobile1.0.4+7, no publication or R07. See Spec177 resume.md/deployment-review.md. No100% claim until required live tasks close.

R06 full activation delegation —2026-11-06: the owner approved the complete Spec177 deployment-review.md packet with “yes go go go go go” and “get it to100% yes ?”. Foundation/runtime/DNS/TLS/Sonnet/all-six-direction/native clients and bounded retirement are delegated. Do not repeat the same permission requests. Preserve exact state/plans, unknown outcomes, R01/R04a/R05 and existing retirement schedules. Fixed R06 expiryNov8 15:19:35Z and release/R07 hold remain. GAIT7a196046.

R06 / Spec177 technical closure — 2026-11-06

This supersedes the24/35preparation checkpoint above. Full SDD and35/35tasks are complete for three independent ordinary OpenClaw/Sonnet organizations, cloud A/B plus local C. All six directions and actual browser/editor/iOS simulator flows, refusals, budget/consent boundaries and original-result loss/restart recovery pass. Exact native evidence counts16subnet calls,3workspace reads,38provider responses; no device command or unauthorized execution. The final Mobile cold-load race is fixed with shared in-flight loading;499Flutter tests and native same-turn recovery pass. Source/runtime hashes and historical failures are in Spec177 verification.

Preserve private roadmap-r06-20261106 state/receipts and never replay spent plans or requests. Two owned cloud Borders/local C are intentionally retained within Nov8window:15:09:35Zquiesce/export,15:14:35Zexact34-resource teardown plus DNS/key, 15:19:35ZGCESTOP/guard fallback. Loaded Mac jobcom.netclaw.r06-retire-48h needs the Mac awake/online; future deletion is not an observed zero-residue result. R05Nov7 schedule and original BorderDecember1scope are unchanged. Source1.18.6/extension 1.6.1/Mobile1.0.4+7 stay unchanged. No source integration, release/public rollout, existing-client upgrade or R07 is authorized by technical closure.

R06 final history qualification: actual cold native CLI listing hit20.25s while model discovery took24.20s. Cloud host renderer opts into bounded60s history reads; composed saved-history proxy deadline130s. Final HUD520passes, cloud22passes. Original browser result recovered with zero provider replay; transcript hashes remain unchanged (38responses). Four-request proxy admission can return429 during initial dashboard polling; existing Refresh chats recovers the read. Preserve these witnesses and exact v4 amendments in Spec177/history-recovery.

R06 acceptance and source integration — 2026-11-06

After the35/35technical handoff, the owner explicitly requested all completed R06 changes committed, a PR merged into main, completed-branch deletion and return to main, plus a short completion write-up and R07–R10 brief. This accepts R06 and supersedes its source-integration hold. Preserve the separate open 177-defenseclaw-sidecar-identity PR by full path; do not merge that unrelated PR or rename historical Spec177. Source integration does not redeploy the frozen qualified runtimes. Existing R05/R06 expiry and cleanup authority are preserved. The roadmap version/release/public rollout hold remains; R07–R10 briefing does not start implementation or authorize new cloud resources. See Spec177 source-integration.md and docs/roadmap/issue-310/R07-R10-next-steps.md.

R07 / Spec178 execution — 2026-11-06

Owner started R07 using saved handoff and directed100%. Full SDD and three agents in separate worktrees authorized; coordinator serializes shared runtime work. Credit exact R03–R06 evidence and close uncovered gaps. Source/isolated native local verification active; no fresh cloud resource/retention extension implied. Preserve R05Nov7/R06Nov8retirement and original BorderDecember1. No version, publication or R08. See specs/178-recovery-gate/verification.md and tasks.md.

R07 technical qualification — 2026-11-06

Spec178 source, local runtime and native browser/Code1.109/iOS simulator recovery qualification are complete; final human timing and roadmap acceptance remain T030. Source branch178-recovery-gate, integrated code32f87526. Federation1756pass /5existing skips, HUD529pass serial, editor209pass/3existing skips, Mobile518pass, builds/analyzer and clean-source coherence pass. New actual pending-Border restart retains original unknown without replay. The new native subnet delivery drill made9 calls across three fresh runs; broader native fixtures are separate. Hosted-provider/device calls0. Native client workloads are controlled; prior R03–R06 cloud credit remains tied to frozen sources.

All owned test services stopped; private simulator shut down and evidence exported. R05/R06 retirement checks/jobs/hashes unchanged; future cleanup unobserved. An early Code1.145 isolated launch touched existing sharedStorage/state.vscdb mtime; no prehash means content impact unknown. Successful native editor proof is1.109; do not claim all owner files untouched. See Spec178 verification.md for limits. No R08, source push/PR/merge, version bump or publication.

R07 acceptance and source integration — 2026-11-07

After the completed evidence/objective review, the owner requested allR07work committed, PRmerged into main, completed-branch cleanup and return to main. This accepts31/31includingT030 and supersedesR07source-integration hold. Keep source 1.18.6/editor1.6.1/Mobile1.0.4+7 and whole-roadmap publicationhold. Preserve unrelated PR332 and frozen runtime qualification. Evaluate R08 together afterward; this authorizes planning discussion only, no R08implementation/cloud activation. R05/R06retirement and originalBorderDec1 unchanged. See Spec178source-integration.

R08 / Spec179 execution — 2026-11-07

Owner directed R08 to100% using full SDD after joint strategy review and explicitly reaffirmed NCFED grounding. This supersedes prior R08 discussion-only hold for source/isolated local qualification. Use NCFED cluster profile and existing iN2N/ eN2N trust boundaries;private etcd only coordinates authority. Three voters, member/guard fresh claims,fixture then actual guarded model/nativeMCP and real HUD/Code/Mobile evidence are required. No R09/cloud activation,existing-runtime upgrade or publication implied. Preserve source1.18.7/editor1.6.1/Mobile1.0.4+7, R05/R06retirement and originalBorderDec1. See specs/179-border-ha-authority/.

R08 implementation team — 2026-11-07

The owner explicitly chose a coordinator and three implementation agents in separate worktrees for R08/Spec179. Root coordinates core service/NCFED authority and serializes all runtime qualification mutations. Parallel source lanes cover lifecycle/recovery, HUD/VS Code and Mobile/enrollment. This does not authorize R09, cloud deployment, publication, version changes or mutation of retained runtimes.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.