memnyx

Authoritative, version-pinned reference for every Claude Code CLI flag, classified by whether it works in an INTERACTIVE session (default TUI), a HEADLESS session (-p/--print), or BOTH — plus the canonical launch recipes for each mode. Use when any session needs to launch or script `claude` and must know which switch is valid in which mode, build a headless/streaming invocation, set up remote-control or an always-on session, pass model/effort/permissions/MCP/plugin/settings flags, or answer "does flag X work with -p / in interactive". Trigger: "claude flags", "claude cli switches", "which flags work headless vs interactive", "-p flags", "stream-json flags", "how do I launch claude with X", "/claude-cli-flags".

Authoritative, version-pinned reference for every built-in Claude Code slash (/) command — what each does, its aliases, and which are bundled Skills/Workflows vs fixed built-ins — plus how custom slash commands work (the commands-merged-into-skills model, file locations, frontmatter, arguments, !bash, @file). Includes the full /loop vs /schedule breakdown. Use when any session needs to know what a slash command does, find the right command for a task (clear context, switch model, review code, schedule work, background a task), or author a custom /command. Trigger: "claude slash commands", "what does /X do", "list the slash commands", "built-in commands", "custom slash command", "/loop vs /schedule", "how do I make a /command", "/claude-slash-commands". Sibling of claude-cli-flags (launch flags).

Initialise a workspace, register projects under it, or sync upstream changes. Single entry point for workspace lifecycle.

Enforce coordinated integration testing — identify tool dependencies, check running systems, test all 4 levels (unit/component/integration/system), never disrupt workflows

Autonomous task loop — runs a task repeatedly until completion criteria are met. No babysitting required.

Dual-model review — runs the same question past a second, independent reasoner (a separate model/CLI, e.g. Codex) and synthesizes the disagreements with the primary Claude take. Use before an irreversible call — architecture, security, strategy, high-blast-radius code. Trigger phrases "second opinion", "/second-opinion", "review this", "get another perspective", "sanity check this plan".

Turn message text into TTS-ready text — strip markdown (so a TTS engine never speaks "asterisk asterisk"), and in voice mode also drop emoji. A single shared cleaner meant to be reused by every channel that speaks text aloud (a local speak command, a chat-bot voice bridge, etc). Use whenever text is about to be synthesized to audio, or whenever a channel renders raw markdown.

Rewrite text so it reads like a person wrote it, not an LLM — strip the AI "tells" (em-dash overuse, "it's not just X, it's Y", rule-of-three everywhere, corporate diction like delve/leverage/robust, reflexive hedging, mechanical transitions and summary-closers, uniform sentence rhythm) while preserving meaning and precision. Register-aware (Slack/peer, internal email, external email, formal doc, marketing). Use when the user says "humanise this", "humanize", "make it sound human", "remove the AI tells", "de-AI this", "make it sound less like a bot", "/humanise", or asks why a draft reads machine-generated.

Persistent task list that survives across sessions, with staleness detection, priority decay, and blocked detection. Markdown-only — no MCP dependency. Canonical file at `<workspace>/.claude/memory/todo/TODO.md`. Use when the user says "add todo", "mark done", "show todos", "stale tasks", "blocked items", "store this as a todo", "overdue" — OR any persistent tracking need (the built-in TaskCreate is session-scoped, this is for things that outlive the session).

Drive a real web browser through cmux's built-in snapshot-driven browser automation — open/navigate, read via accessibility snapshots, click/fill/select forms, attach files, log in via cookies/profile import, screenshot, and submit. Use this instead of a permission-gated browser MCP for filling and submitting web forms (job applications, ATS portals, dashboards) — cmux has no per-domain permission wall, which is what typically blocks that kind of MCP. Trigger on "use the cmux browser", "drive the browser", "fill this form", "submit this application in the browser", "cmux browse", or any task that needs autonomous web-form interaction. Works the same way for any assistant driving the shared cmux CLI, Claude Code or otherwise.

Deterministic playbook for controlling and introspecting cmux terminal tabs and the Claude Code sessions running in them — without re-researching each time. Covers enumerating/indexing every tab, reading & changing tab names, injecting text/commands into a tab from outside, reading a tab's output (read-screen/capture-pane/pipe-pane), launching & driving Claude Code, and inspecting Claude internals (per-tab turn-state, surface→session-id map, on-disk transcript, Feed/approval cards). Also covers the always-on, externally-streamable Claude session pattern. Use when any session needs to list/find tabs, send keystrokes into another tab, read what another tab shows, drive or monitor another Claude Code session, build a fleet monitor, or wire an external command stream into a live session. Trigger: "control cmux", "inject into a tab", "read another tab", "list/index the tabs", "what claude sessions are running", "drive another claude", "send keys to surface", "cmux session map", "/cmux-control".

Install and configure the managed AWS MCP Server for Claude Code

Manage skills — add, update, remove, and review based on lessons or user requests

End the session — summarize work, persist context, capture lessons, close the active marker

Start a session — locate workspace, classify cwd, resolve project + workstream + open item, load memory, write session marker, recap context

Capture and integrate lessons learned. Two modes — capture (single lesson) and scan (mine session files or extended JSONL logs for skill-change proposals).

Definitive reference for how Claude Code works — disambiguates skills vs hooks vs subagents vs MCPs vs slash commands vs memory vs settings. Use when asked "how does Claude Code X work", "what's the difference between X and Y", "where should this live", "build me a [skill|hook|agent|mcp|slash command]", "configure Claude Code", or when picking the right surface for a new capability.

Scribe — the project-documentation skill. Generate or maintain a project's Claude-facing documentation (CLAUDE.md, .claude/docs/*, project-context.md, README) from verified facts. Use when the user says "document this project", "write the docs for X", "the docs are stale", "fill in the .claude docs", or "/scribe". Detects existing doc state and routes between from-scratch and update. Verify-before-assert is the core discipline — confirmed facts go to committed docs, inferred/uncertain ones go to a separate hazards artifact, never silently into the repo.

Speak content aloud via Kokoro neural TTS (local, offline). Use when the user says "read it for me", "read it to me", "play it in audio", "say it", "speak it", "read that aloud", "/say-it", or asks to hear something spoken instead of reading it.

Install a local neural voice interface for Claude Code on macOS Apple Silicon. Wires mlx-whisper (STT) + Kokoro TTS (offline neural voices) into voice-claude and vtranscribe CLI scripts. Two voice contexts — personal (af_heart) and tech (af_bella). Multi-session safe — concurrent sessions speak in turn (global lock) and announce their name. No cloud APIs, no API keys.

Collect the user's daily work activity from Slack, Jira, Confluence, GitHub, and Google Drive with source links

Collect a team member's daily work activity from Slack, Jira, Confluence, and GitHub

Scrub skills, SKILL.md, CLAUDE.md, memory files, and markdown before public/boilerplate release. Detects secrets, PII, private context, and reputation risks. Two-phase — detect → approve → apply. Has a --check mode for CI gates. Called by /contribute and /pull-contributions as the dedicated scrubber.

Audits the Claude Code setup (CLAUDE.md, skills, MCPs) for cost and context efficiency without touching quality. Produces a report with sized, prioritized actions and delegates execution to skills-manager or asks the user for approval. Use weekly, when sessions feel slow, or when bills spike.

Pull generalized contributions from a project into Memnyx. Runs `sanitizer --check` as a mandatory gate before integration.

Install and configure cognee-mcp on this machine

Run the full security analysis pipeline — AWS Inspector V2 + GitHub security alerts → correlation → dashboard. Saves dated snapshots for trend tracking. Run monthly or on demand. Optional --deploy flag pushes the dashboard to Google Apps Script at the end.

Generalize a lesson and stage it for Memnyx contribution. Delegates sanitization to `sanitizer` before staging.

Sweep the user's unread LinkedIn DMs and handle cold outreach in two modes — deflect overt sales pitches with an assertive door-closing reply, and socially probe fresh ambiguous openers with a question that demands a genuine answer (real peers answer, sales personas pivot → then deflect); stay social through the sender's 3rd message, then hand genuine threads to the user. Drives the real logged-in Chrome via the chrome-control MCP. Trigger on "clear my sales pitches", "deflect pitches", "linkedin sales sweep", "kill the incoming sales", "check my linkedin messages", "/linkedin-pitch-deflector".

Manage the workspace's project registry at `<workspace>/.claude/projects-index.json`. Single mutation point for the registry — other skills read the file directly, all writes funnel through here. Actions: `add`, `remove`, `update`, `list`.

Wire the optional auto-memory system into the user's ~/.claude/. Creates the per-project memory directory, copies templates and examples, installs the memory-hygiene curation skill, and offers a CLAUDE.md fragment that documents the type system on disk. Trigger "set up auto-memory", "enable auto-memory", "wire in auto-memory", "/setup-auto-memory".

Deploy an HTML file as a Google Apps Script web app with a stable URL. Called by other skills (e.g. a dashboard-generating skill) with a sourceDir argument. Handles clasp setup, authentication, project creation, and updates. Config lives in sourceDir/clasp-projects.json — no global state.

Unified research skill with three depth modes. Use `--shallow` (single-pass parallel web search via the `research-expert` agent) for "quick research", "look up", "find me info on". Use `--standard` (decompose → parallel subagents → synthesize → cite-check) for "research X", "investigate". Use `--deep` (full 9-stage pipeline with breadth scavenging, depth iteration, gap-fill, contradiction detection, theory drafting, fact-check, three-tier output) for "deep research", "thorough investigation", "full literature review", "research this rigorously". Replaces the older `deep-research-orchestrator` skill.

Install and configure a local Wikibase Suite for projects that need a Wikidata-style knowledge graph with claim-level provenance

Audit and prune the auto-memory system. Classifies each MEMORY.md entry and topic file as ENFORCE / HOT / WARM / STALE, archives old ones, and optionally promotes stable HOT patterns into CLAUDE.md Hard Overrides. Use when MEMORY.md gets dense (>30 entries), sessions feel slow from context bloat, or on a monthly cadence. Trigger "memory hygiene", "clean memory", "memory audit", "memory promote", "prune memory".

Prepare 1:1 meeting notes by synthesizing a team member's recent activity into a structured meeting agenda

Append a structured entry to the agent log with automatic rotation

Check health of configured MCP servers

Install and configure Playwright MCP for browser automation