# /deploy-agents - Launch Specialized Review Agents

Deploy one or more review agents against the current codebase or recent changes.

## Available Agents

{{AGENT_LIST}}

Default agent list (customize per project):

| Agent | Purpose | When to Use |
|-------|---------|-------------|
| `tech-lead` | Architectural guidance | Before starting complex features |
| `code-reviewer` | Code quality review | After implementing features |
| `security-reviewer` | Security + privacy audit | Before merging PRs, before deploy |

## Usage

Ask which agents to deploy. Default: `code-reviewer` + `security-reviewer` for post-implementation review.

### Deploy Pattern

When spawning agents, ALWAYS include:
1. The absolute project path in the prompt
2. Instruction to read CLAUDE.md first
3. The specific review scope (see below)
4. Output destination: `.claude/reports/<agent-name>-<date>.md`
5. "Do NOT commit, push, or modify source files."

```
Use the Agent tool with:
- Load the agent prompt from .claude/agents/<agent-name>.md
- Append the scope context and output instructions
- Run agents in parallel (they write to separate report files)
```

## Review Scope

Determine the right scope before deploying:

```bash
# Changes since last merge to main
git diff origin/main..HEAD --stat

# Or changes in last N commits
git diff HEAD~5 --stat

# Or specific files
git diff --name-only origin/main..HEAD
```

Use `origin/main..HEAD` by default (all changes on the current branch). Fall back to `HEAD~5` if on main.

## Multi-Agent Review

For comprehensive review, deploy in parallel:
1. `code-reviewer` — code quality
2. `security-reviewer` — security + privacy

Agents write to separate report files in `.claude/reports/`. After all agents complete:

1. Read each report from `.claude/reports/`
2. Deduplicate findings (same file:line from multiple agents)
3. Present a unified summary to the user, sorted by severity