stark-skills

Mark a logical phase completion in haivemind. More formal than /snapshot — used at step/phase boundaries in multi-phase work. Creates a named, resumable point that /continue can restore from. Triggered by /checkpoint, "checkpoint this", "mark this as a checkpoint", or at phase boundaries in long-running tasks.

Run the Qwen Code CLI as a one-shot bounded executor in a plan/do/verify loop. A frontier model (or you) plans the next bounded step and verifies the result; qwen-do makes a local model do exactly that one step, smoke-test first, then stop. Use when you want a cheap local model to do grunt coding while a smarter orchestrator keeps it on rails.

Scan git commits for PII, credentials, API keys, and passwords before they are pushed. Automatically gates commits via a pre-commit hook. Use when installing PII scanning on a repo, manually scanning a file or diff for credentials, or when a commit is blocked by the hook and remediation is needed. Triggers on "scan for PII", "check for credentials", "install PII guard", "pii hook", or when a commit is blocked by a PII check.

Runs a full CORE-EEAT 80-item content quality audit, scoring content across 8 dimensions with weighted scoring by content type. Produces a detailed report with per-item scores, dimension analysis, and a prioritized action plan.

Deep analysis and debugging with Opus. Systematically investigate problems, identify root causes, and recommend fixes with evidence.

PR review skill. Two modes: /pr [link] = internal thoughts only, what to look for, what to test — NO GitHub action ever. /pr-comment [link] "make it so #1" = draft a GitHub PR comment in <USER_NAME>s voice, present for approval, post only on explicit "go". Passphrase + "go" are both required before any GitHub write. Without both gates, nothing leaves the building.

Deep reflection and optimization with Opus. Two modes — /self-reflect analyzes completed work, extracts root causes, and stores lessons. /optimize analyzes a process or system, determines the right model and approach, then executes improvements. Use when reflecting on what happened, extracting lessons, optimizing workflows, or improving approaches. Triggers on "self-reflect", "reflect on", "optimize", "what went wrong", "what can we improve", "lessons learned".

Deep static analysis and bug hunting for any codebase.

Call Claude (your existing Claude Code login — whichever account/host) to produce a plan, with optional custom system prompt. Default system prompt is the /plan skill (research-first, design-primer-aware, Qwen-ready format, recommends an executor model). Uses the local `claude` CLI so no API key juggling. Use when user says "/cplan", "claude plan", "cplan with system prompt", or wants Claude-as-planner without setting up keys.

Create the engineering deployment-window meeting in Google Calendar — default 9-11 PM EDT today, default attendees are the engineering team (4 members) plus anyone listed in the latest "<YYYY-MM-DD> — Deploy Worksheet" in Drive. Adds Google Meet, opens link on <MAC_HOST>. Use when user says "create a deployment window", "deploy meeting", "deployment window meeting", or "schedule deploy".

Daily status check-in for a tracked cohort (accounts, services, sensors, anything you watch on a daily cadence). Dispatches a background subagent that pulls per-member status from a configured data source, appends a new dated tab to an output doc, and saves a memory snapshot. Trigger phrases — "/daily-cohort-checkin", "daily checkin", "run cohort status", "checkin <name>".

Standalone Google Doc update step — appends a daily entry (as a new dated tab with real GDocs tables) from data stored in a memory provider. Companion to /daily-cohort-checkin; usable independently whenever you need to retry just the doc-write step without re-running the data gathering. Trigger phrases — "/append-daily-doc-tab", "update daily doc", "retry doc tab", "write today's tab".

Use when switching the local Docker Model Runner model on <MAC_HOST> (:12434). Handles loading/unloading models, checking what's running, and configuring Cline/opencode to use a specific local model. Only one model can run at a time on Metal.

Run long-term memory consolidation pass (full pass + decay-based pruning). Heavy ~30 min run. Trigger phrases: /dream-lt, dream long, lt dream, deep consolidate, weekly memory cleanup

Run short-term memory consolidation pass (delta-only, no pruning). Use after a session ends, or when memories feel stale. Trigger phrases: /dream-st, run dream short, dream-short, st dream, consolidate recent memories

Show dreaming system status: last ST/LT run timestamps, qdrant index health, store divergence, pending consolidations, recent prune log. Use when user says: /dream-status, dream status, how is dreaming, dream health, dreaming system status, check dream health

Use to pause / resume Jarvis voice services (Whisper STT, Chatterbox TTS, Kokoro TTS docker) on <INFERENCE_HOST> to free RTX 4090 VRAM for testing larger Ollama models. Voice will not work while paused — explicit cost. Pairs with the <VOICE_RUNTIME> env-var toggles for durable on/off. Trigger phrases — "/gpu-free", "free up the gpu", "pause jarvis", "stop voice services so I can test a model", "resume jarvis".

Access hAIveMind memory system via MCP tools for search and storage

Memory-backed session handoff. Post a structured snapshot to Notion+hAIveMind so the next session (any machine) can pick up instantly. Replaces Discord-relay handoff. Use when handing work between machines, times, or Claude sessions.

Use when user asks for the Cline Kanban passcode, password, access code, or asks to restart kanban. Pulls the latest passcode straight from the systemd journal so it's always current.

Create a Cline Kanban implementation ticket that picks the right Anthropic model (Sonnet 4.6 by default, escalate to Opus 4.7 for architecture/security/judgment-heavy work) and reasoning effort based on the task content. Works in any git-repo project (<PROJECT_NAME>, <VOICE_RUNTIME>, <PRODUCT_NAME>, anywhere) — auto-registers the project on first ticket if not already registered. Use when the user says "kb-do", "kanban do", "create a kanban ticket to do X", "make a kanban implementation task", or when adding any non-planning Cline Kanban ticket. Sonnet 4.6 or Opus 4.7 only.

Link two Cline Kanban tasks so one waits on the other (auto-cascade pipeline). When a task in review is trashed, any linked backlog task auto-starts. Use when the user says "kb-link", "link these tasks", "make X depend on Y", "chain these tickets", or wants to wire up a sequential pipeline of work. Project-agnostic.

Create a new Cline Kanban project from scratch — `git init` a new repo if needed, drop a starter README.md, make the initial commit, and register it as a Cline Kanban workspace. Works in any directory under ~/Dev/ (or wherever the user specifies). Uses `--no-verify` on the initial commit only if <PROJECT_NAME> pre-commit hooks block (most new empty projects have nothing to flag, so it shouldn't come up). Use when the user says "kb-new", "new kanban project", "make a new project", "create a project for X", or wants to start a fresh Cline Kanban workspace.

Create a Cline Kanban planning ticket that always runs on Opus in plan mode with xhigh reasoning, regardless of the calling agent's model. Works in any git-repo project (<PROJECT_NAME>, <VOICE_RUNTIME>, <PRODUCT_NAME>, anywhere) — auto-registers the project on first ticket if not already registered. Use when the user says "plan this in kanban", "kb-plan", "kanban plan", "write a plan ticket", or wants design/architecture work decomposed before implementation. The ticket produces a written plan (saved as a markdown file in the project) that a follow-up implementation ticket executes.

Show the current Cline Kanban board state for a project — tasks grouped by column (in_progress, review, backlog, trash), with session state, model assignments, and dependency edges. Use when the user says "kb-status", "kanban status", "show the board", "what's on kanban", "what tasks do I have", or wants to see what's running/queued. Project-agnostic — defaults to current directory, accepts `--project <path>`.

Move a Cline Kanban task (or an entire column) to trash, stopping any active session and cleaning up its worktree. Auto-cascades any linked backlog tasks (one in trash → next in chain auto-starts). Use when the user says "kb-trash", "trash that task", "trash kanban", "clean up the review column", "drop those tickets", or wants to remove tasks from the board.

Unified LM Studio control on Mac — pull, load (with optional max context), warm prefill cache, unload, fresh-reload (unload-then-load), status, list. Use when the user says "warm up <model>", "preload", "prime the cache", "load <model>", "unload", "reload <model>", "fresh load", "switch to <model>", "download <model>", "what's loaded", "max context".

Load context from Obsidian memory + Notion (meetings, plans). Use when the user asks to "load", "retrieve", "recall", "check history", or "what do you know about X".

Retro a recorded meeting (transcript source = Notion page or local transcript file) to surface communication patterns, decision quality, follow-up clarity, and reusable lessons. Saves a single retro per meeting and supports a `patterns` cross-meeting synthesis. Invoke as /meeting-retro (latest), /meeting-retro <date or search query>, or /meeting-retro patterns.

Passive live-meeting helper. Polls a Notion meeting-notes page (transcript) every minute via cron, surfaces only deltas, and proposes <USER_NAME>voice answers when a question or decision appears. Optionally watches a Google Doc's comment stream alongside. <USER_NAME>drives the meeting; this skill watches and feeds. Never auto-edits the doc. Use when <USER_NAME>says "/meeting-workshop", "monitor this meeting", "watch this transcript", "live meeting helper", or pastes a Notion meeting-notes URL during a call.

Batch-migrate memories from a semantic memory store (e.g., hAIveMind, or any provider with a `search_memories` MCP tool) into an Obsidian vault as canonical local-first markdown files. One batch per run — designed for cron execution. Reads/writes `.migration-state.json` in vault root for resumability.

Control the MiniMax-M2.7 JANGTQ vmlx server on <MAC_HOST> (:8765). Best local coder <USER_NAME>owns (78% SWE-bench Verified) but holds ~60 GB Mac memory continuously. Use to start, stop, status-check, restart, tail logs, or smoke-test the server. Trigger phrases — "/minimax", "start minimax", "stop minimax", "minimax status", "is minimax loaded", "free minimax memory".

Use when switching the local MLX inference server model on Mac, checking what model is loaded, monitoring downloads, or configuring Cline/Roo Code to use a local model.

Use when switching the loaded Ollama model on <MAC_HOST> (:11434), checking what's loaded, listing available models, or setting PI's default model. Replaces the docker-switch workflow now that Ollama is the primary local-inference backend. Trigger phrases — "/model-switch", "switch ollama model", "load qwen", "load nemotron", "what's loaded on <MAC_HOST>", "set pi default to".

Create a new Notion AI meeting notes page and open it in Chrome on <WORKSTATION_HOST>. Use when user says "new meeting", "notion meeting", "start a meeting", "open notion meeting".

Use when running Ollama commands on <MAC_HOST> () — listing/pulling/showing/running models, inspecting templates, switching context. Captures the right SSH invocation, binary location, env vars, and the gotchas that have bitten before. Pair with /docker-switch for Docker Model Runner side-by-side.

Opus PLAN — produce a structured implementation plan using main Claude (Opus, full context, all tools) in the same format /pplan emits. Output is saved as a markdown plan file that /pdo can execute locally. Use when planning needs cloud-tier reasoning OR full repo context, but execution should stay private. Trigger phrases — "/oplan", "opus plan", "plan with opus", "claude plan and pi do".

Call `claude -p --model opus` to produce an execution-ready plan via the Claude Code CLI. Uses cloud Opus (separate billing) with a system prompt optimized for plans that guide local execution end-to-end. Trigger phrases — "/opusplan", "opus plan via claude", "plan with claude opus".

Pair-editing ritual for customer deliverables. <USER_NAME>watches the live surface (Google Doc, Notion page) in Claude in Chrome; Claude makes edits via MCP (mcporter on <INFERENCE_HOST>). One edit at a time, plan-first, stop after each for visual review. Use when <USER_NAME>says "let's pair edit", "/pair-edit", "edit one at a time", or after pulling comments on a deliverable Doc.

Private DO — delegate a bounded coding/text task to local PI + Ollama. Code stays on the LAN. Spawns a verbose Discord thread by default so progress is visible while you work elsewhere. Trigger phrases — "/pdo", "private do", "private agent", "delegate to pi", "run on local", "let pi do this".