> ## Documentation Index
> Fetch the complete documentation index at: https://docs.useorgx.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Skills

> Reusable instructions that help compatible AI clients follow OrgX workflows and produce reviewable work.

Skills are reusable instruction packs for compatible AI clients. They explain
how to gather context, use the OrgX tools available to that client, respect
decision boundaries, and leave a reviewable work record.

<Info>
  A skill is guidance, not authorization. MCP scopes, workspace permissions,
  provider permissions, budgets, and decision policy still control what a client
  or agent can do.
</Info>

## How skills work

When installed, a skill gives the client a repeatable workflow for a class of
work. A useful skill normally describes:

* the trigger and intended outcome;
* the context to inspect before acting;
* the OrgX tools or provider actions it may use;
* the evidence, artifact, decision, or handoff expected at the end;
* the conditions that require a question, permission, or approval.

Skills do not grant access, replace MCP authentication, or prove that an
external action succeeded. The current [MCP tool catalog](/docs/api/mcp-tools) and
the connected provider contract are authoritative for callable operations.

## Install the starter packs

The wizard can install the default OrgX skill packs into a supported client:

```bash theme={"dark"}
npx @useorgx/wizard@latest skills add
```

To install named packs:

```bash theme={"dark"}
npx @useorgx/wizard@latest skills add engineering-agent initiative-kickoff
```

The exact client configuration path and available pack names can change with
the wizard version. Run the wizard's help command if a pack is not available in
your client.

## Workflow packs

| Pack                  | Use it for                                           | Public result                                                                  |
| --------------------- | ---------------------------------------------------- | ------------------------------------------------------------------------------ |
| `morning-briefing`    | Review pending decisions, blockers, and next actions | A concise review brief based on available workspace records                    |
| `nightly-recap`       | Summarize a work period                              | A bounded recap with unresolved items and evidence links where available       |
| `initiative-kickoff`  | Turn a goal into a reviewable work structure         | A proposed or created hierarchy, subject to client permissions and policy      |
| `initiative-protocol` | Manage initiative lifecycle work                     | Explicit workstream, dependency, and decision checkpoints                      |
| `engineering-agent`   | Technical planning and code review workflows         | A technical artifact or review record with supporting evidence where available |
| `product-agent`       | Product framing and prioritization                   | A product artifact or decision packet                                          |
| `marketing-agent`     | Campaign and content planning                        | A draft artifact and review boundary                                           |
| `sales-agent`         | Outreach and pipeline workflows                      | A draft or reviewable sales artifact                                           |
| `design-agent`        | UX and accessibility review                          | A design brief or assessment                                                   |
| `operations-agent`    | Runbooks, incidents, and operating procedures        | An operational artifact with risks and follow-up actions                       |
| `orchestrator-agent`  | Cross-domain coordination                            | A plan, handoff, or synthesis record                                           |

Availability and behavior depend on the client, workspace, and current tool
catalog. A skill may propose work without executing it.

## Evidence and handoff

Good skill outputs make the next review easy. Where the workflow has evidence,
include a durable reference to the source or external result, identify what was
observed versus inferred, and state what remains unverified. Keep secrets and
raw provider transcripts out of the artifact.

For a client-specific setup path, see the [AI Toolkit](/docs/guides/toolkit) and
[MCP client setup](/docs/guides/cursor-mcp-setup) guides.