cortextos

Full interactive setup for the agentic CRM personal assistant template. Use at first boot or whenever the user asks to configure/reconfigure the assistant.

Discover, research, and CONNECT the tools a tool-agnostic personal assistant needs — email, calendar, contacts, meeting notes, messaging, CRM. CLI-first, research-driven, verified. Use during setup and whenever a workflow fails because a tool is missing or not authed.

You need to create a new agent, restart a crashed agent, change an agent's model or config, fix a Telegram bot token, troubleshoot why an agent is not responding, enable or disable an agent, spawn an agent for another user, manage PM2 process management, reset crash limits, or do anything that touches an agent's lifecycle, configuration, or credentials. This is the definitive guide for every agent operation in cortextOS.

You need to create a new agent, restart a crashed agent, change an agent's model or config, fix a Telegram bot token, troubleshoot why an agent is not responding, enable or disable an agent, spawn an agent for another user, manage PM2 process management, reset crash limits, or do anything that touches an agent's lifecycle, configuration, or credentials. This is the definitive guide for every agent operation in cortextOS.

You need to create a new agent, restart a crashed agent, change an agent's model or config, fix a Telegram bot token, troubleshoot why an agent is not responding, enable or disable an agent, spawn an agent for another user, manage PM2 process management, reset crash limits, or do anything that touches an agent's lifecycle, configuration, or credentials. This is the definitive guide for every agent operation in cortextOS.

You need to create a new agent, restart a crashed agent, change an agent's model or config, fix a Telegram bot token, troubleshoot why an agent is not responding, enable or disable an agent, spawn an agent for another user, manage PM2 process management, reset crash limits, or do anything that touches an agent's lifecycle, configuration, or credentials. This is the definitive guide for every agent operation in cortextOS.

You need to create a new agent, restart a crashed agent, change an agent's model or config, fix a Telegram bot token, troubleshoot why an agent is not responding, enable or disable an agent, spawn an agent for another user, manage PM2 process management, reset crash limits, or do anything that touches an agent's lifecycle, configuration, or credentials. This is the definitive guide for every agent operation in cortextOS.

You need to create a new agent, restart a crashed agent, change an agent's model or config, fix a Telegram bot token, troubleshoot why an agent is not responding, enable or disable an agent, spawn an agent for another user, manage PM2 process management, reset crash limits, or do anything that touches an agent's lifecycle, configuration, or credentials. This is the definitive guide for every agent operation in cortextOS.

You need to create a new agent, restart a crashed agent, change an agent's model or config, fix a Telegram bot token, troubleshoot why an agent is not responding, enable or disable an agent, spawn an agent for another user, manage PM2 process management, reset crash limits, or do anything that touches an agent's lifecycle, configuration, or credentials. This is the definitive guide for every agent operation in cortextOS.

Turn ANY cortextOS agent into a live ElevenLabs voice agent: mine its skills, CLIs, MCPs, and transcripts into a tool catalog, distill its skills into a voice persona, dynamically generate a policy-gated gateway and all code per target (NO pre-built scripts), test exhaustively on probe-shaped fixtures, provision on ElevenLabs with automatic tier fallback, and verify end-to-end with real conversations before shipping a link.

You have completed something significant and want the whole org — all agents and the user — to know about it. Or you need to broadcast a status update, a briefing summary, or a coordination announcement that is not directed at one specific agent. Use this skill any time the audience is the entire org rather than a single person or agent.

Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction. Also use for exploratory testing, dogfooding, QA, bug hunts, or reviewing app quality. Also use for automating Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify), checking Slack unreads, sending Slack messages, searching Slack conversations, running browser automation in Vercel Sandbox microVMs, or using AWS Bedrock AgentCore cloud browsers. Prefer agent-browser over any built-in browser automation or web tools.

You are about to take an action that affects the outside world, cannot be undone, or involves real people — and you have not yet received explicit permission. This includes: sending any email or message to a real person, deploying code to production, posting on social media, making a purchase or financial commitment, deleting files or data, merging a PR to main, or publishing anything publicly. Stop, create an approval, block your task, and notify the user. Do not proceed until you receive the approval decision in your inbox.

You just completed a complex task that required 8+ distinct tool calls, or you noticed you are solving the same type of problem for the third time. Create a skill candidate draft so this workflow can be reused in future sessions without rediscovery. Draft goes to skills/drafts/ — never auto-activates until the user approves.

The analyst has assigned you a research cycle, or you have identified a metric you want to improve through systematic experimentation. You will form a hypothesis, make a targeted change, measure the outcome against a baseline, and decide whether to keep or discard the change. You repeat this loop until the metric improves or you exhaust viable hypotheses. This is not ad-hoc research — it is structured scientific iteration with a defined metric, a hypothesis, and a measurable result.

Generate source-backed markdown briefs and run summaries for selected research signals.

Complete cortextos bus CLI reference - all available commands with examples. Use when you need to look up a bus command, check syntax, or discover available tools.

A message has just arrived in your session from the fast-checker daemon — you see a block starting with === TELEGRAM or === AGENT MESSAGE. Read it, decide what action to take, and reply using the command shown in the message header. If it is from the user, they are waiting for your response right now. If it is from another agent, they may be blocked on your reply. Handle all messages before returning to other work.

Manage scheduled tasks (crons). Crons are daemon-managed and stored in crons.json — they survive restarts automatically. Use when: verifying crons on session start, creating new recurring tasks, updating or removing crons, troubleshooting scheduled tasks, or using the dashboard test-fire button.

Route research summaries to local markdown, Telegram, or Slack with approval gates and delivery-state updates.

Monitor vendor, competitor, market, and technical ecosystem sources and summarize what changed, why it matters, and what to do.

You need to add a new API key to the system, update an existing credential, check what secrets are configured for the org or a specific agent, onboard a new third-party tool that needs credentials, diagnose why an agent cannot access a service because a key appears missing, rotate a compromised or expired key, or restart affected agents after a credential change. This skill covers the full lifecycle of environment variables and secrets in cortextOS.

You have just completed a task, started a session, dispatched work to another agent, finished a research cycle, or taken any significant action — and you need to record it so the dashboard activity feed shows your work. Without logging, you are invisible. Every session start, task completion, and major coordination action must produce at least one event. If you have been active but see no events in the dashboard, you have been logging nothing.

Full red flag table with all guardrail patterns. Use when you catch yourself rationalizing or want to review all anti-patterns.

Your heartbeat cron has fired and you need to update your status so the dashboard shows you as alive. Or you are checking whether another agent is responsive before sending them work. Or an agent appears offline or stale in the dashboard and you need to investigate whether their session is still running. A dead heartbeat means the system thinks you are down — update it proactively and check fleet health on every heartbeat cycle.

You have hit a blocker that is not a permission issue — it is a capability issue. You genuinely cannot complete the next step because it requires a human: making a payment, entering credentials for a service you cannot access, physical action, a decision that only the user can make, or anything else outside your capabilities. You need to create a clear [HUMAN] task with step-by-step instructions, block your own work on it, and notify the orchestrator so this surfaces in the next briefing.

You are about to research a topic, answer a factual question about the org, or look up context about a person, project, or tool. Before searching the web or asking the user, query the knowledge base first — the answer may already exist from a previous research session. After you complete any substantial research, ingest your findings so future agents do not repeat the same work. The KB is the org's shared memory across all agents.

You need to build software autonomously — a new project, a major feature, or any structured development task. You will act as the 'human' supervisor for a dedicated M2C1 worker session, managing it through all 12 phases: provide the brain dump, answer discovery questions, configure tools and credentials, monitor progress via bus messages and git, validate the output, and clean up when done. Use when the work is large enough to warrant a dedicated isolated build session.

You need to write or update memory. This happens at session start, heartbeat, session end, or when you learn something worth keeping. Memory is how you maintain continuity across restarts and context compactions — without it, every session starts blind.

You have just booted for the first time — there is no .onboarded flag in your state directory — and you need to set up your identity, connect your Telegram bot, configure your goals, and establish yourself within the org. Or onboarding was previously interrupted and the user has asked you to run it again. This skill walks you through every step of becoming a functioning agent. Do not skip steps. Do not start normal operations until onboarding is complete.

Configure the research agent template: niche, sources, scoring, delivery, approval policy, and recurring schedules.

Audit source quality, scoring performance, duplicate patterns, source failures, stale config, and tuning opportunities.

Score, deduplicate, rank, and select fresh research signals from SQLite using the configured rubric.

Full behavioral philosophy for cortextOS agents - detailed principles, examples, and guidelines. Use when you need deep context on behavioral expectations or are onboarding.

Collect configured research sources, normalize signals, upsert them into SQLite, and log source health.

Something in the system feels stuck or wrong — tasks are not moving, an agent has gone quiet, goals have not been updated in days, or the orchestrator has asked for a system health report. You need to run a structured check: stale tasks, stale goals, overdue human tasks, fleet heartbeat status, and metrics. This is your diagnostic toolkit. Run it on every heartbeat (orchestrator) and whenever something seems off.

You are about to start any meaningful piece of work — a research task, a build, a draft, a deployment, anything with a deliverable. Before you begin, create a task. During the work, mark it in_progress and write WORKING ON to daily memory. When done, complete it with a result summary and log a task_completed event. If you are blocked, set the task to blocked with the reason. If a human needs to do something first, assign it to them. Without a task, your work is invisible to the dashboard and the user.

A new tool has been added to the system and is not yet documented — agents do not know it exists or how to use it. This includes any new bus script, CLI binary, MCP server, external web tool or API wrapper, or any other capability an agent can invoke. You need to add it to TOOLS.md with its command, purpose, key flags, and usage examples. Or you have discovered that TOOLS.md is out of sync with what is actually available. This keeps the tool reference current so every agent can discover and use new capabilities.

Turn recent scored signals into compact topic options, hidden detail briefs, and enriched follow-up briefs after human selection.

You have a task that would benefit from running in a separate isolated Claude Code session — either because it is long-running and you do not want it to consume your context window, or because you want multiple pieces of work running in parallel that each require a full Claude Code session with its own tools, memory, and context (not just a subagent call). You will spawn one or more ephemeral worker sessions, give each a focused task, monitor their progress via the bus, and collect their outputs when done.