Agents, Skills, and Scheduled Tasks

The two Base agents

• RapidClaw is the product orchestrator and general operating agent. It coordinates work through the configured agents, skills, connectors, tools, models, and policies.
• Arcy is the Command Center guide, builder, diagnostician, and governed operator assistant. It helps administrators change the operating environment safely.

Agent Roster distinguishes RapidClaw-managed agents from custom agents created for your organization. Product extensions may add managed specialist agents, but they use the same shared runtime and governance path.

Create or edit an agent with Arcy

1. Open Arcy and start the guided agent builder, or select an existing agent to edit.
2. Describe the outcome, responsibilities, boundaries, and expected audience in ordinary language.
3. Choose from models actually deployed in the environment. Arcy can recommend an appropriate available model for the workload.
4. Select only the connectors, tools, skills, and data sources the agent needs.
5. Review the generated instructions, scope, permission requirements, and approval boundary.
6. Save the agent as a draft or approve activation when the configuration is ready.

Build reusable skills

Skills package repeatable instructions and capabilities that can be used by the appropriate agents. Use Skills Catalog to review existing skills and Arcy's guided skill builder to create a new one. Define the input, expected output, intended agents, data boundary, tools, and safety requirements. A skill should be narrow enough to test and review independently.

• Use explicit business outcomes instead of vague goals.
• Keep read-only research separate from actions that write, post, send, or delete.
• Attach only the tools required for the skill.
• Test with representative but non-sensitive examples before broader use.
• Update the skill rather than copying slightly different versions into many agents.

Create scheduled tasks

A scheduled task tells an agent when to perform defined recurring work. Arcy can draft the prompt, cadence, agent, model, data scope, tools, and delivery target using live configuration selectors.

1. Choose the agent and state the recurring business outcome.
2. Define the schedule, time zone, and expected delivery destination.
3. Select the required connectors and tools and review their permission boundaries.
4. Confirm whether the task only prepares information or could request a protected action.
5. Save the schedule disabled, test its output, and activate it only after review.

Review changes over time

• Use Agent Instructions and Permission Rules to understand the current operating boundary.
• Use Activity Log and Agent Journal to review what ran and how agents behaved.
• Use Usage & Costs to compare model consumption by agent, including Arcy.
• Pause or disable a schedule when its output, scope, or timing no longer matches the business need.
• Use Emergency Stop when governed execution must be frozen across agents, schedules, and Teams while diagnosis continues through Arcy.