PMOVES.AI

Operate the P7 room-aware stage manager — read a room's current stage, transition a room (gated rehearsal→live via the canonical CHIT activation checklist), and observe lifecycle events on NATS. Use when claiming/transitioning a room, when the operator asks "what stage is room X in?", or when wiring a UI to reflect room transitions in real-time.

Email ingestion, triage, and review workflows for Titanmail and Gmail accounts using the Himalaya CLI and local-only processing.

Folder-triggered document processing workflows with zero-retention E2B sandbox execution for legal, research, and operations pipelines.

Legal document review, research, and case-law verification for PMOVES.AI tenant-advocacy and contract analysis workloads.

Activates n8n and Archon as the local workflow and agent orchestration layer for email, research, verification, and legal-assist jobs.

YouTube and SoundCloud channel monitoring with transcript extraction, batch processing, and routing to Hi-RAG or notebooks.

Multi-target deployment bridge for PMOVES.AI infrastructure. Manages local Docker Compose, Kubernetes AI Lab, KVM4 production, VPS fleet deployment, homelab bare-metal, and edge (Jetson) nodes from a single Pinokio control panel. Built-in gepeto + pinokio skills. Covers: Z890, 5090, 4090, SPARK, R9700, JONS×3, KVM4-1/4-2, KVM2.

Generate CGP (Creative Gamespace Packet) content for PMOVES creator workflows on the KiloCode GLM lane. Use when generating CGP packets, creating creative content, or executing creator demos.

Execute hyperdimensional operations for PMOVES BoTZ tactical partner dynamics on the KiloCode GLM lane. Use when performing high-dimensional computations, tensor operations, or operating in the Transformers/Bumblebee energy framework.

Verify tool execution using text, audio, and VLM checks across PMOVES services on the KiloCode GLM lane. Use when validating multimodal outputs, capturing verification evidence, or binding verification to task/run IDs.

Ensure PMOVES overlays align with upstream submodule capabilities on the KiloCode GLM lane. Use when auditing git submodule alignment, checking overlay compatibility, or validating integration contracts.

Transform source materials into grounded persona anchors for PMOVES agents on the KiloCode GLM lane. Use when creating personas, grounding agent behavior, or establishing persona policy metadata.

Map secrets stores into CHIT manifests without cleartext commits for PMOVES infrastructure on the KiloCode GLM lane. Use when syncing GitHub secrets, vault labels, or Supabase runtime values to CHIT manifests.

Execute wave-function collapse operations for hyperdimensional state space navigation in PMOVES on the KiloCode GLM lane. Use when performing quantum-inspired pathfinding, state collapse operations, or wave-function navigation.

Execute AGENT TRAILS roguelike visualization framework for PMOVES agent navigation on the KiloCode GLM lane. Use when visualizing agent execution lanes, navigating time crystals, or operating in the Transformers/Bumblebee energy framework.

Guide for tiered bring-up, smoke validation, and evidence capture during PMOVES service deployments on the KiloCode GLM lane. Use when starting services, running smoke tests, or validating CI/CD pipeline outputs.

Blueprint-first feature implementation for PMOVES.AI using GLM-5.2 coding plan. Use when implementing features from field briefs, architecture specs, or design documents with a structured blueprint approach.

Sign Graphiti trail entries with DARKXSIDE COCREATOR witness attribution for PMOVES.AI. Use when completing work items, signing trail entries, or performing agent handoffs on the 5090 node.

Fleet status reporting and GPU mesh coordination for PMOVES.AI on the 5090 node. Use when checking service health, GPU utilization, fleet connectivity, or coordinating model load/unload events.

MCP server development and integration for PMOVES.AI. Use when building MCP servers, configuring MCP clients, or wiring agent-to-agent communication via Model Context Protocol.

Pull and register Ollama models on the 5090 GPU node, sync with TensorZero config, and manage model availability for the PMOVES fleet. Use when adding new local models or updating model routing.

Ensure KiloCode configuration parity with Claude Code and Codex across PMOVES.AI submodules. Use when auditing config consistency, checking claw scope alignment, or verifying that all three 5090 agents have equivalent capabilities.

Launch an AgentGym RL training session on this node using the 4090 field runner config (pmoves/configs/agentgym/field-runner-4090.yaml). Agent model: Qwen 3.5 9B via TensorZero (localhost:3030). Publishes episode results to agentgym.episode.completed.v1 on NATS. Environments: BabyAI, TextCraft, Maze, Wordle (lightweight); ALFWorld, SQLGym, WebShop (moderate). Max 300s/15 rounds, concurrency=1. Shift Crew style: NATS event publishing after each episode.

Launch the PMOVES Demo Room: starts Agent Zero locally, loads 4090 Claude Code claws (skills), and surfaces HERMES V4 as the reasoning assist. Publishes p7.nats.launch to trigger the P7 room-aware stage manager. Use /demo-room to enter demo mode from any node.

Template a KiloCode implementation brief at .kilo/command/<name>.md. Generates the required sections: ## Arguments, ## Implementation, ## Related, ## Notes. Uses Three-Body pattern: Claude analyzes → KiloCode (GLM-5.1 on 5090) implements → trail signed on completion.

Pull local inference models (Hermes V4, Qwen3 embed, Gemma 4 multimodal, NeMo Omni, Unsloth GGUF variants) via Ollama or HuggingFace CLI. Node-aware routing: ≥70B → SPARK only; 8B → any node with Ollama; embedding models → Z890/RDNA4/Spark; multimodal (text/image/audio in, text out) → Spark/5090/4090; NeMo Omni → Spark only. Use /model-pull <model-id> or describe what you want to pull.

Run the W0 Substrate hardware probe on the 4090 node. Windows: deploy/provision/glances-autodetect.ps1 Linux: deploy/provision/glances-autodetect.sh Outputs structured JSON: gpu, cpu, nics, nic_collisions, unifi_topology, system specs. Pipe through json-to-profile.py to write pmoves/config/profiles/<node>.yaml.

Full situation report for the 4090 laptop node: git status, active worktrees, GPU (nvidia-smi), NATS health (port 8222), running containers, and open PRs. Output formatted as AGNOTE4482 audit record block. Run before every CLAIM/RELEASE in AGNOTE4482PHI.t1.md.

Verify the 4090 node's W0 substrate artifacts landed correctly on main. Checks: node-4090-probe SKILL.md has no BLOCKED notices, json-to-profile.py has unifi_topology wired, TAC probe-wire is status:done. Run after any W0 PR merge. For secrets health, use deploy:secrets-funnel.

Bind a FlOO$ character suit to the session voice pipeline. Sets BEATS_VOICE env var and documents CGP control_plane.param_surface overrides for: dr-bean (analytical/slow), mr-clean (command/crisp), powerpuff-bubbles (coordination), powerpuff-blossom (joy), powerpuff-buttercup (action).

Enable persistent memory across Claude Code sessions for PMOVES.AI development.

Single-shot beats-to-voice pipeline. Encodes text at a target BPM into a CGP v0.2 prosodic packet and publishes to tokenism.prosodic.bpm.v1. BPM values: 60 (sentence/largo), 80 (breath/adagio), 90 (clause/andante), 120 (phrase/moderato), 150 (none/presto). Use shift-listen for daemon mode.

Start beats_to_voice.py in NATS push-listen mode on this node. Subscribes to voice.agent.response.v1, publishes CGP v0.2 prosodic packets to tokenism.prosodic.bpm.v1. Daemon — runs until stopped. Requires: NATS reachable, beats_to_voice.py available.

PMOVES-KIMI session bootstrap and model/agent selection guidance for Kimi Code CLI operators.

Sign a CHIT trail payload, append a CLAIM-or-RELEASE entry to AGNOTE4482PHI.t1.md, and stage a NATS publish to chit.signed.v1 (via pmoves-nats-mcp when wired). Use when committing CHIT-aware service changes.

Measure what a PMOVES exit node buys a household vs its raw local uplink (A/B), and map participating homes onto measured node capacity. Use for the Fordham Hill community-mesh pilot, exit-node speed/selection checks, and capacity planning.

Run /healthz across every service in .claude/CATALOG.md and emit pass/fail. Use before claiming work in AGNOTE4482PHI.t1.md.

Drive the real ComfyUI UI to run the Ideogram-Ultra image workflow via the chrome-devtools MCP, narrating each knob to teach the user, and harvest the POST /prompt API-format payload. Used by the creator-operator (L1). The end user never touches the node graph.

Sync PMOVES forks from upstream and promote PMOVES.AI submodule gitlinks, driven by the PMOVES.AI GitHub App. Use when forks are behind upstream (CVE/feature drift), when "the submodules aren't synced", or before a gitlink-promotion commit. Covers the App token Known Road, the two-layer sync (fork←upstream, gitlink←fork), and the manual CRITICAL merge.

Diagnose PMOVES GHCR/CI failures by SIGNATURE before guessing a cause — startup_failure (YAML break / Actions allowlist) vs disk-no-space vs cancel vs real-CVE. Turns "CI is red again" into a 30-second triage with the matching fix. Use when a GHCR publish or any workflow run goes red/startup_failure.