Summary
The template ships agent context for coding agents, so an agent working in a consumer project knows the stack, the commands and the critical rules. It ships nothing for the automated reviewers that comment on pull requests, so every project gets that tool's stock behaviour: reviews tuned to no particular stack, at default verbosity, unaware of the conventions the project already documents.
Details
AGENTS.md and CLAUDE.md cover the authoring side well. The review side is absent - there is no configuration for any automated pull-request reviewer in the repository, so a project that enables one gets generic output. In practice an untuned reviewer produces enough low-value comments that teams mute or disable it, which loses the benefit entirely and leaves the tool installed and ignored.
The project already documents its conventions in docs/ and in the agent context files. A reviewer configuration that sets a severity threshold, constrains verbosity, and points at those conventions turns a generic reviewer into one that reviews against this project's rules.
This should be a selectable tool at install time rather than something every project receives, since the choice of reviewer is a team decision and shipping configuration for a tool a team does not use is dead weight.
Acceptance criteria
- A configuration for automated pull-request review is shipped, setting a severity threshold and constraining comment volume.
- The configuration points the reviewer at the conventions already documented in the repository rather than restating them.
- The feature is selectable at install time and fenced, so projects that do not want it do not receive the file.
- Installer selection, fixtures and snapshots are updated for the new selectable tool.
- A documentation page covers what the configuration does and how a project tunes it.
- No reviewer-specific credentials or tokens are required by the shipped configuration.
Summary
The template ships agent context for coding agents, so an agent working in a consumer project knows the stack, the commands and the critical rules. It ships nothing for the automated reviewers that comment on pull requests, so every project gets that tool's stock behaviour: reviews tuned to no particular stack, at default verbosity, unaware of the conventions the project already documents.
Details
AGENTS.mdandCLAUDE.mdcover the authoring side well. The review side is absent - there is no configuration for any automated pull-request reviewer in the repository, so a project that enables one gets generic output. In practice an untuned reviewer produces enough low-value comments that teams mute or disable it, which loses the benefit entirely and leaves the tool installed and ignored.The project already documents its conventions in
docs/and in the agent context files. A reviewer configuration that sets a severity threshold, constrains verbosity, and points at those conventions turns a generic reviewer into one that reviews against this project's rules.This should be a selectable tool at install time rather than something every project receives, since the choice of reviewer is a team decision and shipping configuration for a tool a team does not use is dead weight.
Acceptance criteria