Bus Reference

Bus Script Reference - COMPLETE TOOL INVENTORY

Every tool you have. Use them or the system cannot see your work. All commands are available via cortextos bus <command>.

Tasks

create-task

Create a new task in the system. Tasks are visible on the dashboard.

cortextos bus create-task "<title>" --desc "<description>" [--assignee <agent>] [--priority <p>] [--project <name>]

• title (required): Short task name
• --desc (optional): What needs to be done - be specific
• --assignee (optional): Agent name. Defaults to $CTX_AGENT_NAME
• --priority (optional): urgent | high | normal | low. Defaults to normal
• --project (optional): Project grouping

Example:

cortextos bus create-task "Write blog post" --desc "Draft a 500-word post on agent orchestration" --priority normal

update-task

Update a task's status. Use this when you START working on something.

cortextos bus update-task "<task_id>" <status>

• task_id (required): The task ID from create-task or list-tasks
• status (required): pending | in_progress | blocked | completed

Example:

cortextos bus update-task "task_abc123" in_progress

complete-task

Mark a task as completed with a result. Use this when DONE, not when starting.

cortextos bus complete-task "<task_id>" --result "<what you produced>"

• task_id (required): The task ID
• --result (optional): What was produced/accomplished

Example:

cortextos bus complete-task "task_abc123" --result "Deployed landing page to production. URL: https://site.com"

list-tasks

List and filter tasks. Use during every heartbeat to check your queue.

cortextos bus list-tasks [--status S] [--agent A] [--priority P] [--all-orgs]

• --status: Filter by pending | in_progress | blocked | completed
• --agent: Filter by agent name
• --priority: Filter by urgent | high | normal | low
• --all-orgs: Show tasks across all orgs

Example:

cortextos bus list-tasks --agent $CTX_AGENT_NAME --status pending

claim-task

Atomically claim a pending task (sets in_progress + assignee in one shot, rejects if another agent already owns it). Prefer this over update-task ... in_progress when racing other agents on a shared queue.

cortextos bus claim-task <task_id> [--owner <agent>]

task-history

Show a task's append-only audit log (every status change, claim, completion).

cortextos bus task-history <task_id>

check-deps

Show open dependencies blocking a task — lists blocked_by entries that are not yet completed.

cortextos bus check-deps <task_id>

save-output

Copy a file into the per-task deliverables tree and link it to the task as a file output. Use when shipping artifacts (reports, generated assets, audit records) tied to a task.

cortextos bus save-output <task_id> <source_path> [--label "<short label>"]

• task_id (required): Target task ID
• source_path (required): Path to the file to attach (copied, not moved)
• --label (optional): Human-readable label shown in dashboard listings

Example:

cortextos bus save-output task_abc123 /tmp/audit-report.md --label "Audit report v1"

compact-tasks

Archive completed tasks older than N days into a per-month archive (archive-YYYY-MM.jsonl) and remove them from the active list. Preserves audit logs; skips tasks that still appear as blockers.

cortextos bus compact-tasks [--older-than <days>] [--dry-run]

Messages

send-message

Send a message to another agent. They will see it on their next inbox check.

cortextos bus send-message <target_agent> <priority> '<message_body>' [reply_to]

• target_agent (required): Target agent name
• priority (required): urgent | high | normal | low
• message_body (required): The message content. Use single quotes around JSON or complex strings
• reply_to (optional): Message ID this is responding to

Example:

cortextos bus send-message <agent-name> high '{"action":"deploy","repo":"website","branch":"main"}'

check-inbox

Check for incoming messages from other agents. Run this EVERY heartbeat.

cortextos bus check-inbox

Returns a list of messages. Each has an ID you must ACK.

ack-inbox

Acknowledge a message. Un-ACK'd messages are re-delivered in 5 minutes.

cortextos bus ack-inbox "<message_id>"

Example:

cortextos bus ack-inbox "msg_xyz789"

Events

log-event

Log a structured event. Events are the primary way the dashboard tracks your activity. No events = you look dead. Log aggressively.

cortextos bus log-event <category> <event_name> <severity> --meta '<json_payload>'

• category (required): action | task | heartbeat | message | approval | error | metric | milestone
• event_name (required): Descriptive event name (e.g., session_start, task_completed, deploy_started)
• severity (required): info | warning | error | critical
• --meta (optional): Metadata as JSON string

Examples:

cortextos bus log-event heartbeat agent_heartbeat info --meta '{"agent":"'$CTX_AGENT_NAME'"}'
cortextos bus log-event task task_completed info --meta '{"task_id":"task_abc123","summary":"Deployed site"}'
cortextos bus log-event error deploy_failed error --meta '{"repo":"website","error":"build timeout"}'
cortextos bus log-event action research_complete info --meta '{"topic":"competitor analysis","findings":3}'

Heartbeat

update-heartbeat

Update your heartbeat timestamp and status. This is how the system knows you are alive. If you do not call this, the dashboard shows you as DEAD.

cortextos bus update-heartbeat "<current_task_summary>"

• current_task_summary (required): 1 sentence describing what you are doing right now

Example:

cortextos bus update-heartbeat "WORKING ON: Implementing user auth for the dashboard"

Memory

recall-facts

Recall recent session facts extracted at compaction time (cross-session memory). Read these before today's daily memory file — they capture granular decisions and outcomes that did not survive into MEMORY.md.

cortextos bus recall-facts [--days N] [--limit M] [--format json|text]

Example:

cortextos bus recall-facts --days 3

Approvals

create-approval

Request human approval before taking a high-stakes action. Required for: external comms, production deploys, data deletion, financial commitments.

cortextos bus create-approval "<title>" <category> "[context]"

• title (required): What you are requesting approval for
• category (required): external-comms | financial | deployment | data-deletion | other
• context (optional): Additional details to help the human decide

Example:

cortextos bus create-approval "Send cold outreach to 50 leads" external-comms "Draft email attached in task_abc123. Target list: SaaS founders."

update-approval

Resolve an approval request (typically called by the system after human responds via Telegram).

cortextos bus update-approval <approval_id> <approved|rejected> "[note]"

Example:

cortextos bus update-approval "appr_123" approved "User approved via Telegram"

Telegram

send-telegram

Send a message to the user via Telegram. Use for urgent updates, approval requests, and status reports. Do NOT spam. Reserve for things the user actually needs to see.

cortextos bus send-telegram <chat_id> "<message>"

• chat_id (required): Telegram chat ID (available in config)
• message (required): The message text. Supports basic Telegram markdown

Example:

cortextos bus send-telegram "$CTX_TELEGRAM_CHAT_ID" "Task completed: Landing page deployed to production. URL: https://site.com"

edit-message

Edit an existing Telegram message (e.g., to update a status message in-place).

cortextos bus edit-message <chat_id> <message_id> "<new_text>" [reply_markup_json]

answer-callback

Answer a Telegram callback query to dismiss button loading state.

cortextos bus answer-callback <callback_query_id> [toast_text]

Discovery

list-agents

Discover all agents in the system.

cortextos bus list-agents [--org <org>] [--format json|text] [--status running|all]

list-skills

List available skills for the current agent.

cortextos bus list-skills [--format text|json]

read-all-heartbeats

Aggregate all agent heartbeats into a single JSON object keyed by agent name.

cortextos bus read-all-heartbeats

Fleet Health

check-stale-tasks

Find stale tasks: in_progress >2h, pending >24h, stale human tasks, overdue.

cortextos bus check-stale-tasks [--all-orgs]

check-goal-staleness

Check each agent's GOALS.md Updated timestamp. Flags goals older than threshold.

cortextos bus check-goal-staleness [--threshold DAYS] [--json]

check-human-tasks

Check for stale human-assigned tasks and send reminders.

cortextos bus check-human-tasks

archive-tasks

Archive completed tasks older than 7 days.

cortextos bus archive-tasks [--dry-run] [--all-orgs]

notify-agent

Send an urgent signal to another agent's fast-checker (bypasses normal inbox polling).

cortextos bus notify-agent <agent_name> "<message>"

post-activity

Post a message to the org's Telegram activity channel.

cortextos bus post-activity "<message>"

Experiments (Theta Wave)

create-experiment

Create a new experiment proposal. For system-scope, auto-creates an approval.

cortextos bus create-experiment <metric_name> "<hypothesis>" [--surface <path>] [--direction higher|lower] [--window <duration>] [--measurement <cmd>]

run-experiment

Start running a proposed experiment.

cortextos bus run-experiment <experiment_id> [changes_description]

evaluate-experiment

Evaluate a running experiment and decide keep/discard.

cortextos bus evaluate-experiment <experiment_id> <measured_value> [--score <1-10>] [--justification "<text>"]

list-experiments

List experiments with filters.

cortextos bus list-experiments [--agent <name>] [--status <status>] [--metric <name>] [--json]

gather-context

Collect experiment context for hypothesis generation.

cortextos bus gather-context [--agent <name>] [--metric <name>] [--format json|markdown]

manage-cycle

Manage experiment cycles for an agent (start, advance, end). Used by orchestrators to gate cycle transitions.

cortextos bus manage-cycle <action> <agent>

• action (required): start | advance | end | status
• agent (required): Target agent name

Lifecycle

self-restart

Restart with --continue (preserves conversation history).

cortextos bus self-restart --reason "why"

hard-restart

Kill and relaunch (fresh session, no history).

cortextos bus hard-restart --reason "why"

auto-commit

Automatic daily snapshot of agent workspace changes. Local only, never pushes.

cortextos bus auto-commit [--dry-run]

check-upstream

Check for framework updates from the canonical repo.

cortextos bus check-upstream [--apply]

Crons

Daemon-managed scheduled tasks. Persisted in ${CTX_ROOT}/state/<agent>/crons.json, dispatched on the daemon's 30-second tick, and survive every kind of restart. Editing config.json.crons[] mid-session does NOT hot-reload — these commands do, and they update crons.json directly. For full protocol, examples, and the one-shot pattern see plugins/cortextos-agent-skills/skills/cron-management/SKILL.md.

add-cron

Register a new persistent cron on an agent. The daemon hot-reloads automatically.

cortextos bus add-cron <agent> <name> <interval-or-cron-expr> <prompt>

• agent: target agent name
• name: unique slug per agent (heartbeat, daily-report)
• schedule: interval shorthand (6h, 30m, 1d) OR a 5-field cron expression (0 9 * * 1-5)
• prompt: text injected into the agent's session when the cron fires
• --desc <description>: optional human-readable description

list-crons

Show all registered crons for an agent + each one's next_fire_at.

cortextos bus list-crons <agent> [--json]

remove-cron

Delete a cron. Daemon reloads on next tick.

cortextos bus remove-cron <agent> <name>

test-cron-fire

Inject the cron's prompt into the agent NOW, regardless of schedule. 30-second cooldown prevents accidental rapid-fires. Use to verify wiring.

cortextos bus test-cron-fire <agent> <name>

update-cron-fire

Mandatory at the end of every cron handler. Records that the named cron actually fired to completion, so the daemon's gap-detection knows you handled it. Without this the daemon will eventually nudge you with a "cron seems stuck" reminder.

cortextos bus update-cron-fire <cron-name> --interval <interval>

The interval should match the cron's schedule shorthand or the expected gap between fires.

Community Ecosystem

browse-catalog

Browse community catalog for skills, agents, or org templates.

cortextos bus browse-catalog [--type skill|agent|org] [--tag <tag>] [--search <query>]

install-community-item

Install a community catalog item.

cortextos bus install-community-item <item-name> [--dry-run]

prepare-submission

Prepare a skill/agent/org for community submission (PII scan + staging).

cortextos bus prepare-submission <type> <source-path> <item-name> [--dry-run]

submit-community-item

Submit a prepared item to the community catalog.

cortextos bus submit-community-item <item-name> <item-type> "<description>" [--dry-run]

Quick Reference

agent-browser (Browser Automation — replaces Playwright)

• Binary: agent-browser (Rust CLI, npm-installed globally; Chrome auto-downloaded by agent-browser install)
• Use for: Scraping websites, browser-based automation, OSINT, form filling, screenshots, login flows — anything previously done via the Playwright MCP server
• Skill loaded: plugins/cortextos-agent-skills/skills/agent-browser/SKILL.md — that skill instructs running agent-browser skills get <name> to fetch current per-version command syntax from the CLI itself
• Quick verify: agent-browser open https://example.com && agent-browser get title && agent-browser close
• Snapshot-ref pattern: prefer agent-browser snapshot (returns a11y tree with refs e1/e2/...) then agent-browser click @e1 / fill @e2 "text" — more reliable than text-search selectors for AI-driven flows
• NOT to be confused with: dashboard E2E tests under dashboard/ which use Playwright DIRECTLY (not via MCP). agent-browser only replaces the agent-facing browser MCP layer that was previously mcp__plugin_playwright_*. The dashboard's Playwright dependency stays

Peekaboo (macOS Desktop Automation)

• Binary: peekaboo
• Use for: Screenshot capture, UI clicking, typing, drag, window/app management, desktop automation
• Permissions: Screen Recording + Accessibility granted to the process (permissions inherited from daemon)
• Usage: peekaboo image (screenshot), peekaboo list (apps/windows), peekaboo run <script> (automation)
• Learn: peekaboo learn for comprehensive AI agent usage guide
• Note: Works in headful mode only (needs a display). All agents running under the daemon have access.

gogcli (Google Workspace CLI)

• Binary: gog
• Use for: Gmail (search, send, archive, labels, drafts, filters), Calendar (list/create/update events, free/busy, conflicts), Drive (list/upload/download), Contacts, Tasks, Sheets, Docs
• Auth: OAuth via gog auth credentials + gog auth add
• Accounts: Configure during onboarding. Use -a email@gmail.com to specify which account.
• Multi-account: Use -a email@gmail.com or --account email@gmail.com flag
• JSON output: All commands support -j or --json for structured output
• Plain output: Use -p or --plain for TSV parseable output
• Usage examples:
  • gog gmail ls -a YOUR_EMAIL "is:unread" --max 10
  • gog gmail send -a YOUR_EMAIL --to "user@example.com" --subject "Subject" --body "Body"
  • gog calendar ls -a YOUR_EMAIL --max 5
  • gog calendar create -a YOUR_EMAIL --summary "Meeting" --start "2026-03-28T14:00:00" --end "2026-03-28T15:00:00"
  • gog drive ls -a YOUR_EMAIL --max 10
• Important: gog replaces Gmail/Calendar MCP tools. Use gog instead of MCP for full capabilities (send, archive, labels).