How to use the Agent-Operated Software agent skill

What the skill makes the agent do

The skill gives an agent a playbook for an application whose functioning runtime includes an OpenRig agent or rig in its live backend or control loop. The application surface can be thin: structured state and specialist agent roles do the work, and the UI makes that work visible and steerable.

It sets out a shared vocabulary that separates three ideas: AI-enabled software (application code owns the control loop and calls a model as one capability), Agent-Operated Workflows (an agent owns one bounded procedure with a start and stop condition), and Agent-Operated Software (agents take part in an ongoing application's runtime).

• Keep behavior in Markdown, YAML, or JSON control files, with intent, work state, evidence, and decisions in stable, addressed artifacts.
• Make generated artifacts verifiable: write to a partial path, verify the property that can be wrong, flush and rename atomically, then write a manifest and a .done sentinel last.
• Fix problems in the source every agent reads, such as the generator, skill, or schema, rather than in a rendered or installed copy.
• Trace the path that produced a surprising behavior before deciding which layer to repair.

Install the skill from the ZIP or the Claude apps

Download the skill ZIP from the directory page. You then have two ways to add it.

For Claude Code, unzip the package into ~/.claude/skills/ to make it available for you, or into a project's .claude/skills/ folder to limit it to that project.

In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/
• Project install: the .claude/skills/ folder inside your project
• App install: Settings → Capabilities → Skills

Trigger the skill with natural requests

You do not need to name the skill. The skill's description tells the agent to load it when you are designing, building, operating, or diagnosing an ongoing application whose live backend or control loop includes OpenRig agents. That includes apps that keep their control plane in Markdown, YAML, or JSON, and apps that put a thin surface over specialist agent roles.

Describe the application and the problem in plain terms. Name OpenRig and the layer you suspect if you know it.

• Help me design an app where specialist OpenRig agents run the backend and a simple UI shows their progress.
• Our agent-driven app behaved strangely after yesterday's change. Work out which layer caused it.
• Review how we make generated reports verifiable before other agents read them.
• Where should I fix this: the skill file or the copy the agent installed?

What the skill needs and what it does not do

The skill requires OpenRig. Its guidance refers to the rig CLI and the openrig-operating-model reference, and it names four sibling skills: agent-operated-workflows, forming-an-openrig-mental-model, mission-slice-sop, and openrig-user. The reviewed file contains only SKILL.md, so those siblings are not included in this package. Where the skill points to them, you will need them installed separately, or the agent will have less to work with.

The skill runs no scripts and makes no network calls, according to its listing. It is guidance, not a tool that acts on your system.

• Requires: OpenRig, the rig CLI, and the openrig-operating-model reference
• Named but not included: agent-operated-workflows, forming-an-openrig-mental-model, mission-slice-sop, openrig-user
• Runs scripts: no
• Uses network: no

Safety notes on scripts and network access

Because the skill runs no scripts and uses no network, its direct risk is low. Its influence is still real: it changes how an agent decides what to edit and where to fix a problem. Review the changes an agent proposes to control files, schemas, and skills before accepting them.

The skill also draws a line you should respect. Core OpenRig substrate bugs need full reproduction and code fixes, with the full gate held. The looser coherence-first approach applies only to recoverable application layers whose state can be inspected and repaired cheaply. Do not let the agent apply the faster posture to core infrastructure.

• Review proposed edits to control files and skills before you accept them.
• Do not apply the coherence-first approach to OpenRig core bugs.
• Keep historical locks and evidence addressable rather than overwriting them.

Tips for getting good results

Ask the agent to trace the path that produced a behavior before it edits anything. The skill's central advice is that a surprising result may be a code defect or a gap in the control plane, and the fix depends on which.

For studio-style apps, the skill suggests the running product is usually the best place to iterate. It recommends a separate mockup only when it resolves a real design question the running app cannot answer as cheaply.

• State whether the problem is in OpenRig core or in your application layer.
• Ask for fixes in the authoritative source, not in an installed copy.
• Use the sibling skills and rig proof --help for the supported judgment and read path, as the skill directs.