manage-commands

Manage Commands

Commands package multi-step investigations into one named operation. Users invoke them in workspace chat as /command-name. Every engineer gets the same quality investigation without crafting perfect prompts.

When to use

• The same diagnostic sequence is repeated by multiple engineers
• "Tribal" investigation steps are known only by senior responders
• Variable quality between responders for the same issue type
• User asks to create an "investigation", "runbook", "procedure", or "workflow" for chat
• Onboarding new engineers who need guided entry points

Key concepts

Scoping

Scope	Effect	scope_type	scope_id
Workspace	All users in the workspace can invoke	workspace	workspace name
Persona	Only available when chatting with a specific assistant	persona	persona name

Command vs Task

Concept	What it is	Invoked by
Command	Instructions for the AI assistant — what to investigate, how to present results	Users typing /name in chat
Task (SLX)	A script that runs on a runner against live infrastructure	run_slx or the platform scheduler

Commands can reference Tasks. Instead of hardcoding diagnostic details, a command can instruct the assistant to run or consult a Task for live data. This keeps commands current without manual updates.

Naming rules

Command names must be alphanumeric, underscore, or hyphen only. No spaces. Users invoke them as /command-name in workspace chat.

Workflow

1. Audit existing commands

list_chat_commands(scope_type="workspace", scope_id="my-workspace")

2. Identify repeatable patterns

Look for investigation steps your team does repeatedly:

• "Check pods, then events, then logs in namespace X"
• "Compare config between dev and prod"
• "What changed since last deployment?"

3. Create the command

create_chat_command(
    name="investigate-namespace",
    command_content="""Investigate the health of a Kubernetes namespace. Follow this sequence:

1. **Pod status** — Check for crash loops, pending pods, and recent restarts
2. **Warning events** — Surface Kubernetes warning events from the last hour
3. **Error logs** — Check application logs for error patterns
4. **Resource pressure** — Note any resource quota exhaustion
5. **Dependencies** — Check if upstream/downstream services are healthy

Present findings grouped by severity. For each issue found, include specific next steps.
If the namespace is healthy, confirm it explicitly and note any recent recoveries.""",
    scope_type="workspace",
    scope_id="my-workspace",
    description="Comprehensive namespace health investigation"
)