# Valdr Skills
The docs in this section cover the wrappers shipped in `valdr-packs/skills`.

These skills are intentionally small. Each one either resolves the right Valdr agent or loads the right capability prompt, then hands control to that loaded source of truth. That keeps CLI behavior aligned with the current pack configuration instead of baking workflow logic into stale documentation.

> [!IMPORTANT]
> **Valdr Skills require a paid plan.**
> Skills are available on **Vanguard** and **Sovereign** subscriptions with MCP access. The free tier lets you explore the UI — Skills let you work from the CLI without losing structure.

> [!NOTE]
> Valdr ships **five** wrappers: `valdr-orchestrator`, `valdr-planner`, `valdr-executor`, `valdr-reviewer`, and `valdr-auditor`.

## Who Skills Are For

Skills are for developers who prefer working from the terminal but want Valdr's structure and traceability.

- **Claude Code** or **Codex CLI** with MCP support
- **Vanguard** or **Sovereign** subscription
- **Valdr MCP server** running locally

## Available Skills

{{< cards >}}
  {{< card link="valdr-orchestrator/" title="Orchestrator" icon="search" subtitle="Navigate and discover your workspace" >}}
  {{< card link="valdr-planner/" title="Planner" icon="clipboard-list" subtitle="Structure ideas into executable plans" >}}
  {{< card link="valdr-executor/" title="Executor" icon="play" subtitle="Execute tasks with consistent workflow" >}}
  {{< card link="valdr-reviewer/" title="Reviewer" icon="check" subtitle="Quality gates that can't be skipped" >}}
  {{< card link="valdr-auditor/" title="Auditor" icon="eye" subtitle="Evaluate agent work with structured scoring" >}}
{{< /cards >}}

## What Ships Today

| Skill | Wrapper Responsibility | Source of Truth It Loads |
|-------|------------------------|---------------------------|
| **Orchestrator** | General PM discovery, registry operations, and sprint routing | An orchestrator agent prompt resolved from the registry |
| **Planner** | Structured planning, requirements, specs, and task generation | A planner agent prompt resolved from the registry |
| **Executor** | Task execution flow for a specific task key | Executor capabilities plus the assignee prompt from `pm_task.get_prompt` |
| **Reviewer** | Reviewer resolution and review routing for a task | A reviewer agent prompt resolved from task assignments or the registry |
| **Auditor** | Auditor resolution and audit routing for a session | An auditor agent prompt resolved from the registry |

## How Skills Work

Every wrapper follows the same contract:

1. **Resolve** the correct agent, capability, or task-specific prompt.
2. **Load** that prompt or capability from Valdr.
3. **Execute** the loaded workflow as the single source of truth.

That design is deliberate. It keeps the skill wrapper stable while the real behavior lives in prompts and capabilities you can update centrally.

## Shared Protocol

All skill packs inherit the same non-negotiable rules from the pack README:

1. **Fetch before mutate.** Never update tasks, reviews, agents, or sprints blind.
2. **Attach audit metadata on mutations.** Include `actorHandle`, a fresh `clientRequestId`, and a dated reason when the API expects one.

If you change how work should be scored, reviewed, or routed, update the loaded prompt or capability. The wrapper should stay thin.

## Invoke the Current Set

Invoke a wrapper in Claude Code or Codex CLI:

```
/valdr-orchestrator
/valdr-planner
/valdr-executor
/valdr-reviewer
/valdr-auditor
```

## Prompt Gallery

Start with the wrapper slash command, then give it a concrete request:

| Skill | Good first prompt |
|-------|-------------------|
| **Orchestrator** | `"Use gunnar to show open tasks in api-platform and summarize blockers."` |
| **Planner** | `"Use freya to create a plan for adding audit-log export to the admin UI."` |
| **Executor** | `"Execute task WEB-142 and stop at the review gate when the work is ready."` |
| **Reviewer** | `"Review task WEB-142 with reviewer sigrid and focus on risky shell commands."` |
| **Auditor** | `"Use tyr-v2 to audit the latest executor session for WEB-142 and record a score run."` |

For handle-driven wrappers, naming the agent directly in your prompt is the fastest path. If you omit the handle, Valdr asks you which planner, reviewer, auditor, or orchestrator should take the request.

## Installation

Install the pack into your CLI of choice with the Make targets from `valdr-packs`:

```bash
make sync-skills-codex
make sync-skills-claude
make sync-skills
```

`make sync-skills` installs the current wrappers for project-level Claude, Codex, and Gemini setups.

## How the Wrappers Fit Together

```mermaid
flowchart TB
    ORC["🔍 **ORCHESTRATOR**<br/>Discovery, registry, sprints"]
    PLN["📋 **PLANNER**<br/>Plans, specs, requirements"]
    EXE["▶️ **EXECUTOR**<br/>Task execution"]
    REV["✅ **REVIEWER**<br/>Review routing"]
    AUD["🔎 **AUDITOR**<br/>Audit routing"]

    ORC --> PLN
    ORC --> EXE
    ORC --> REV
    ORC --> AUD
    PLN --> EXE
    EXE --> REV
    REV --> AUD
```

## Next Steps

- **[Orchestrator](valdr-orchestrator/)** — Start here for discovery, registry work, and sprint routing
- **[Planner](valdr-planner/)** — Use when work needs a structured plan or task generation
- **[Executor](valdr-executor/)** — Use when a task key exists and the work should begin
- **[Reviewer](valdr-reviewer/)** — Use when a task needs review coverage or a verification gate
- **[Auditor](valdr-auditor/)** — Use when you need a scored evaluation of a session