Custom Agent Tasks

Use roborev run to execute custom tasks with AI agents. While automatic reviews focus on commits, run lets you target specific files, ask questions, or perform targeted analysis.

roborev run "Review src/auth.go for security issues"
roborev run --wait "Find simplification opportunities in this codebase"
roborev run --agentic "Add input validation to the user controller"

Use Cases

Targeted File Reviews

Review specific files for particular concerns:

# Security review
roborev run "Review src/auth/ for security vulnerabilities"
roborev run "Check database queries in src/db/ for SQL injection"

# Simplification opportunities
roborev run "Find opportunities to simplify src/handlers/user.go"
roborev run "Identify dead code or unused functions in src/utils/"

# Refactoring candidates
roborev run "Suggest refactoring opportunities in src/api/"
roborev run "Find code duplication in src/services/"

Code Analysis

Ask questions about architecture and design:

roborev run "Explain the authentication flow in this codebase"
roborev run "What design patterns are used here?"
roborev run "Document the main entry points of this application"
roborev run "List all external API dependencies"

Custom Review Criteria

Apply specific review criteria beyond the default review:

roborev run "Review recent changes for OWASP Top 10 vulnerabilities"
roborev run "Check for proper error handling in src/api/"
roborev run "Verify all database connections are properly closed"
roborev run "Find functions exceeding 50 lines"

Making Changes (Agentic Mode)

Use --agentic to allow file modifications:

roborev run --agentic "Add comprehensive error handling to main.go"
roborev run --agentic "Refactor database layer to use connection pooling"
roborev run --agentic "Add input validation to all API endpoints"
roborev run --agentic "Convert callback-style code to async/await"

Piped Input

Pipe complex prompts or instructions from files:

echo "Add comprehensive error handling" | roborev run --agentic --wait
cat review-checklist.txt | roborev run --wait

Flags

Flag	Description
`--wait`	Wait for task to complete and show result
`--agent`	Agent to use (default: from config)
`--reasoning`	Reasoning level: `fast`, `standard`, or `thorough`
`--no-context`	Don’t include repository context in prompt
`--agentic`	Enable agentic mode (allow file edits and commands)
`--yolo`	Alias for `--agentic`
`--quiet`	Suppress output (just enqueue)

Review vs Agentic Mode

By default, tasks run in review mode (read-only). Use --agentic to enable write operations.

Mode	Tools Available	Use Case
Review (default)	Read, Glob, Grep	Analysis, explanations, finding issues
Agentic	Read, Glob, Grep, Edit, Write, Bash	Refactoring, implementing features, fixes

Repository Context

By default, tasks include context about the repository:

Repository name and path
Any project guidelines from .roborev.toml

Use --no-context for raw prompts without this context.

Tips

Use --wait to see results immediately in the terminal
Use --reasoning thorough for security-sensitive analysis
Combine with roborev tui to review task results later
Tasks appear in the TUI alongside commit reviews