Review Settings

This page is for maintainers tuning the shared review engine and local review-and-fix loop.

Review Engine

Setting	Type and accepted values	Fallback or scaffold	Tier and security note	Example
`prflow_review.verdict_severity_threshold`	`critical`, `important` or `suggestion`	`critical`	Standalone review and review-and-fix review pass. Lowering the threshold makes more findings reject.	`"verdict_severity_threshold": "critical"`
`prflow_review.live_progress_comment_enabled`	Boolean	`true`	Pull-request review. When true, each run maintains its own progress comment. Comment writes are best effort.	`"live_progress_comment_enabled": true`
`prflow_review.stale_prose.enabled`	Boolean	`true`; anything except explicit false enables	Shared review engine. Disabling removes the automatic check for prose that contains stale numeric claims.	`"enabled": true`
`prflow_review.stale_prose.severity`	`critical`, `important` or `suggestion`	`important`	Shared review engine. The chosen severity participates in verdict computation.	`"severity": "important"`

Review and Fix

Setting	Type and accepted values	Fallback or scaffold	Tier and security note	Example
`prflow_review_and_fix.fix_severity_threshold`	`critical`, `important` or `suggestion`	`important`	Local review-and-fix and implementation’s inline fix loop. Findings severe enough to cause a rejection remain eligible for correction at every value.	`"fix_severity_threshold": "important"`
`prflow_review_and_fix.max_iterations`	Integer one or greater	`5`; values below one clamp to one	Review-and-fix. Higher values can increase cost.	`"max_iterations": 5`
`prflow_review_and_fix.efficiency_telemetry_enabled`	Boolean	`true`	Review-and-fix. False disables the effectiveness record and prevents denied-command records from being persisted on the telemetry branch.	`"efficiency_telemetry_enabled": true`
`prflow_review_and_fix.efficiency_cut_candidate_min_dispatch`	Integer one or greater	`3`	Cross-run analysis. It is recorded for later analysis and does not cut an agent during the current run.	`"efficiency_cut_candidate_min_dispatch": 3`
`receiving_review.fix_severity_threshold`	`critical`, `important` or `suggestion`	`critical`	Direct use of the receiving-code-review skill only. The review-and-fix loop uses its own threshold.	`"fix_severity_threshold": "critical"`

Retained Legacy Automatic-Review Settings

Fresh installs do not include automatic pull-request-triggered review. These settings remain only so repositories that installed the withdrawn tier continue to validate. Keep them at their scaffolded values unless you are maintaining such an installation and accept its documented limitations.

Setting	Type and accepted values	Fallback or scaffold	Legacy tier and security note	Example
`prflow_review.require_up_to_date`	Boolean	`true`	Withheld automatic-review caller only. No effect in fresh installs.	`"require_up_to_date": true`
`prflow_review.require_ci_green`	Boolean	`true`	Withheld automatic-review caller only. No effect in fresh installs.	`"require_ci_green": true`
`prflow_review.stall_backstop.enabled`	Boolean	`true`	Withheld automatic-review path and limited manual failure recovery. A resume also needs an App and allowed bot identity.	`"enabled": true`
`prflow_review.stall_backstop.max_resume_attempts`	Integer zero or greater	`2`	Retained review backstop. `0` disables resume while retaining detection.	`"max_resume_attempts": 2`
`prflow_runner.effort`	`low`, `medium`, `high`, `xhigh` or `max`	Scaffold: `low`	Withheld reusable runner. No effect in fresh installs.	`"effort": "low"`
`prflow_runner.provision_env`	Boolean	Runtime false; scaffold: `false`	Withheld runner. Enabling runs pull-request code with provisioned tools under a privileged workflow.	`"provision_env": false`

Provider and allowed-tool settings for prflow_runner are listed in Providers and Tool Permissions, with the same legacy label.

⌘I