Agent Skills

Install slash commands that let AI agents request reviews and fix findings directly:

roborev skills install

Available Skills

Skill	Description
`/roborev:review [commit] [--type ...]`	Request a code review for a commit
`/roborev:review-branch [--base ...] [--type ...]`	Review all commits on the current branch
`/roborev:design-review [commit]`	Request a design review for a commit
`/roborev:design-review-branch [--base ...]`	Design review all commits on the current branch
`/roborev:fix [job_id...]`	Discover and fix all open review findings in one pass
`/roborev:respond <job_id> [message]`	Add a response to document changes

Usage

Review a commit

Request a code review without leaving your agent session:

/roborev:review
/roborev:review abc123
/roborev:review --type security

The review runs in the background via roborev review --wait, so you can keep working. When it completes, the agent presents the findings and offers to address them with /roborev:address.

Review a branch

Review all commits since the current branch diverged from main:

/roborev:review-branch
/roborev:review-branch --base develop
/roborev:review-branch --type security

This runs roborev review --branch --wait in the background. Same workflow: results are presented when ready, with an offer to address findings.

Design review

Request a design-focused review that evaluates completeness, feasibility, and task scoping:

/roborev:design-review
/roborev:design-review abc123

Runs roborev review --wait --type design in the background, following the same pattern as the other review skills.

Design review a branch

Review all commits on the current branch with a design-focused lens:

/roborev:design-review-branch
/roborev:design-review-branch --base develop

This is the branch equivalent of /roborev:design-review, running roborev review --branch --wait --type design in the background.

Fix all open reviews at once

The most powerful skill is /roborev:fix. With no arguments it discovers all open failed reviews on recent commits and fixes them in a single pass:

/roborev:fix

You can also target specific jobs:

/roborev:fix 1019 1021

The agent:

Discovers open reviews (or uses provided job IDs)
Fetches all reviews and collects findings
Groups findings by file and prioritizes by severity
Fixes all issues across all reviews
Runs tests to verify
Records a comment on each closed review
Offers to commit

This is the interactive equivalent of roborev fix --batch — the agent sees all findings at once and can make coordinated fixes across related issues.

Fix a single review

Target a specific job ID with /roborev:fix:

/roborev:fix 1019

The agent fetches the review, fixes issues by priority, runs tests, and offers to commit.

Agent-Specific Syntax

Agent	Syntax
Claude Code	`/roborev:review`, `/roborev:review-branch`, `/roborev:design-review`, `/roborev:design-review-branch`, `/roborev:fix`, `/roborev:respond`
Codex	`$roborev:review`, `$roborev:review-branch`, `$roborev:design-review`, `$roborev:design-review-branch`, `$roborev:fix`, `$roborev:respond`

Checking Skill Status

See which skills are installed and whether any need updating:

roborev skills

Updating Skills

Skills are updated automatically when you run:

roborev update

How It Works

Skills are installed as agent-specific configuration:

Claude Code: Custom slash commands in ~/.claude/
Codex: Custom agent skills directory

The review skills call roborev review --wait to enqueue reviews and wait for results. The fix skills call roborev show --job <id> --json to fetch review data, then parse and present findings to the agent in a structured format.

Waiting for Hook-Triggered Reviews

When a post-commit hook already enqueues reviews, agents don’t need roborev review --wait (which would create a duplicate job). Use roborev wait instead:

git commit -m "Fix auth validation"   # Hook triggers review
roborev wait --quiet                  # Block until verdict (exit 0=pass, 1=fail)

This is more token-efficient than polling roborev list or roborev show because the agent makes a single blocking call and reads the exit code. See Waiting for a Review Without Enqueuing for the full flag reference.

Interactive vs Automated

Skills provide an interactive workflow. The review skills (/roborev:review, /roborev:review-branch, /roborev:design-review, /roborev:design-review-branch) let you request reviews from within your agent session and see results inline. The /roborev:fix skill lets your agent address findings while you review the changes before committing.

For fully automated fixing, use roborev fix --batch (headless, no agent interaction) or roborev refine (iterative loop until all reviews pass).