Skip to content

PR Review Configuration Reference

Complete reference for .rp1/config/pr-review.yaml configuration options.

Overview

The PR review configuration file controls how automated PR reviews behave in CI/CD environments. The file is located at:

.rp1/config/pr-review.yaml

When the configuration file is missing, default values are used. In CI mode with enabled: false (the default), the review is skipped entirely.

Schema

# Enable PR review in CI mode
# When false in CI, review exits early without posting
# Type: boolean
# Default: false
enabled: false

# Review draft PRs
# When false, draft PRs are skipped entirely
# Type: boolean
# Default: true
review_drafts: true

# AI harness for running the review
# Type: "claude-code" | "opencode"
# Default: "claude-code"
ai_harness: claude-code

# Post inline comments on specific code locations
# When false, only a summary review is posted
# Type: boolean
# Default: true
add_comments: true

# Collapse summary in HTML <details> tag
# Useful for reducing visual noise on large reviews
# Type: boolean
# Default: false
collapse_summary: false

# Verdict mode for PR review submission
# Type: "approve" | "request_changes" | "comment" | "auto"
# Default: "auto"
verdict: auto

# Maximum inline comments per review
# Prevents overwhelming PR authors
# Type: integer
# Default: 25
max_comments: 25

# Hidden HTML comment for bot identification
# Used for deduplication across runs
# Type: string
# Default: "<!-- rp1-review -->"
bot_marker: "<!-- rp1-review -->"

# Generate mermaid diagrams in review summary
# Diagrams show component flows and change relationships
# Type: boolean
# Default: false
visualize: false

# CI platform for context extraction
# Type: "github" | "buildkite" | "gitlab"
# Default: "github"
ci_platform: github

Option Details

enabled

Property Value
Type boolean
Default false
Required No

Controls whether PR review runs in CI mode. When false and running in CI, the review exits early without posting any comments or review.

Example: 

enabled: true  # Enable automated reviews

CI Only

This setting only affects CI mode. Local runs (/pr-review in an IDE) always proceed regardless of this setting.

review_drafts

Property Value
Type boolean
Default true
Required No

Controls whether draft PRs are reviewed.

Value Behavior
true Review draft PRs normally
false Skip draft PRs (exit early, no review posted)

Example: 

review_drafts: false  # Don't review work-in-progress

ai_harness

Property Value
Type "claude-code" | "opencode"
Default "claude-code"
Required No

Specifies which AI coding assistant runs the review.

Example: 

ai_harness: opencode  # Use OpenCode instead of Claude Code

add_comments

Property Value
Type boolean
Default true
Required No

Controls whether inline comments are posted on specific code locations.

Value Behavior
true Post inline comments + summary review
false Post summary review only (no inline comments)

Example: 

add_comments: false  # Summary only, less noise

Reducing Noise

Set add_comments: false for large teams where inline comments might be overwhelming. The summary still contains all findings.

collapse_summary

Property Value
Type boolean
Default false
Required No

Wraps the review summary in an HTML <details> tag for collapsible display.

Example: 

collapse_summary: true  # Collapsible summary

Rendered output: 

<details>
<summary>rp1 PR Review Summary</summary>

... full review content ...

</details>

verdict

Property Value
Type "approve" | "request_changes" | "comment" | "auto"
Default "auto"
Required No

Controls the GitHub review verdict submitted with the review.

Mode Behavior Use Case
auto Verdict based on finding severity Most teams
approve Always submit APPROVE Advisory-only reviews
request_changes Always submit REQUEST_CHANGES Strict enforcement
comment Always submit COMMENT Neutral feedback

Auto verdict logic:

Findings Verdict
Critical or High severity REQUEST_CHANGES
Medium or Low severity only COMMENT
No findings APPROVE

Example: 

verdict: comment  # Never block PRs, just provide feedback

max_comments

Property Value
Type integer
Default 25
Required No

Limits the number of inline comments posted per review. When findings exceed this limit, only the highest severity items are posted.

Example: 

max_comments: 10  # Limit to 10 most important comments

Summary Unaffected

The review summary always contains all findings regardless of this limit.

bot_marker

Property Value
Type string
Default "<!-- rp1-review -->"
Required No

Hidden HTML comment embedded in all bot comments for identification. Used for:

  • Deduplication across multiple runs
  • Identifying bot comments vs human comments
  • Analytics and tracking

Example: 

bot_marker: "<!-- my-team-rp1 -->"

Consistency

Changing bot_marker will cause deduplication to fail for existing comments. Only change this when setting up a new repository.

visualize

Property Value
Type boolean
Default false
Required No

Generates mermaid diagrams showing change relationships and component flows.

Value Behavior
true Include mermaid diagrams in review summary
false Text-only review (faster)

Example: 

visualize: true  # Include visual diagrams

GitHub Rendering

GitHub natively renders mermaid code blocks in PR comments. No special setup required.

ci_platform

Property Value
Type "github" | "buildkite" | "gitlab"
Default "github"
Required No

Specifies which CI platform to use for context extraction. This determines how PR metadata is retrieved.

Platform Description
github Use GitHub Actions environment variables and event payload
buildkite Use Buildkite environment variables
gitlab Use GitLab CI environment variables

Example: 

ci_platform: github  # Use GitHub Actions (default)

CI Detection

The CI environment is detected automatically via standard environment variables (CI=true, GITHUB_ACTIONS=true, etc.). The ci_platform setting only controls which platform's context extraction is used.

Environment Variable Overrides

Configuration values can be overridden via environment variables for CI flexibility:

Environment Variable Overrides Example
RP1_PR_REVIEW_ENABLED enabled true
RP1_PR_REVIEW_VERDICT verdict request_changes
RP1_PR_REVIEW_ADD_COMMENTS add_comments false
RP1_PR_REVIEW_VISUALIZE visualize true

Precedence: Environment variables override file configuration.

Example in GitHub Actions: 

- name: Run Strict Review
  env:
    RP1_PR_REVIEW_VERDICT: request_changes
  run: claude /rp1-dev:pr-review

Complete Example

# .rp1/config/pr-review.yaml
#
# Configuration for automated PR reviews in CI/CD

# Enable the feature
enabled: true

# Skip draft PRs - only review when ready
review_drafts: false

# Use Claude Code
ai_harness: claude-code

# Post inline comments for specific findings
add_comments: true

# Don't collapse - show full summary
collapse_summary: false

# Auto-determine verdict based on severity
verdict: auto

# Limit comments to avoid overwhelming authors
max_comments: 20

# Custom marker for this repo
bot_marker: "<!-- rp1-review -->"

# Include visual diagrams for architectural context
visualize: true

# Use GitHub Actions for CI (default)
ci_platform: github

Validation Errors

Invalid configuration produces clear error messages:

Error Cause Fix
Invalid verdict value verdict not one of allowed values Use auto, approve, request_changes, or comment
Invalid ai_harness value ai_harness not recognized Use claude-code or opencode
Invalid ci_platform value ci_platform not recognized Use github, buildkite, or gitlab
max_comments must be positive Negative or zero value Use a positive integer
YAML parse error Invalid YAML syntax Check indentation and structure

See Also