prr-cli v0.1.0

prr-cli 0.1.0

Initial public release of prr-cli, a Node.js CLI for running parallel code reviews and concept brainstorms against OpenAI-compatible language models, including NVIDIA-hosted and OpenAI backends.

Highlights

• Run PR/code reviews from the CLI in either sequential or parallel mode.
• Use separate reviewer lanes for defects and suggestions workflows.
• Let tool-enabled reviewers fetch more file context on demand.
• Use structured outputs with json_schema and json_object modes when supported by the provider.
• Run prr brainstorm <brief.md> to review a concept brief or implementation plan instead of source code.
• Add optional brainstormSynthesis output to turn panel findings into a prioritized recommendation and revised next brief.
• Include auto-selection logic that prefers verified elite tool-capable models.
• Include NVIDIA model probe tooling and experiment helpers for evaluating model/reviewer combinations.

What’s new in this release

Brainstorm mode

This release introduces a first-class brainstorm workflow:

• separate brainstormers config from normal reviewers
• markdown brief input instead of source-file review input
• no file selection, diff parsing, or tool use required for the core brainstorm loop
• optional synthesis stage for producing a tighter recommendation and a reusable markdown nextBrief

This makes it practical to fan out across many low-cost or free NVIDIA-hosted models for concept exploration before implementation starts.

Review workflow improvements

• clearer reviewer-mode split between bug-finding and suggestion-focused lanes
• configurable suggestion caps
• improved GPT-5 guidance for maxOutputTokens and reasoning-token budgeting
• explicit support for leaving maxOutputTokens unset when you want the provider to decide the cap

Reliability and release hardening

• verified elite tool-capable model list and recommendation flow
• packaging tightened with an explicit npm files allowlist
• repository, homepage, and issue tracker metadata included in package metadata
• brainstorm synthesis parsing hardened so nextBrief is preserved correctly in final structured output

Validation

Release validation included:

• npm run lint
• npm test
• npm pack --dry-run
• live OpenAI-compatible request validation
• live brainstorm smoke testing against the configured NVIDIA-backed example flow

Notes

• brainstormSynthesis remains optional, so existing brainstorm runs can work without a synthesis stage.
• For NVIDIA and other non-GPT-5 models, leaving maxOutputTokens unset means prr-cli does not send max_tokens.
• For GPT-5-family models, maxOutputTokens maps to max_completion_tokens, so it should usually be left unset or set conservatively high.