@wes-dev/pr 0.0.1

package/CHANGELOG.md

@@ -0,0 +1,20 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.0.1] - 2026-07-09
+
+### Added
+
+- Local-first pull request review workspace available through `npx @wes-dev/pr`.
+- Explicit pull request targeting with repository references and GitHub pull request URLs.
+- GitHub CLI integration for loading pull requests, reading comments, and submitting reviews.
+- In-memory draft comments and review decisions.
+- LLM-assisted logical grouping using configurable OpenAI-compatible local runtimes.
+- Selectable logical groups with chunk-filtered diffs and file navigation.
+- Local-only server startup, a bundled production CLI, and a pinned Corepack-managed pnpm version.

package/README.md

@@ -0,0 +1,82 @@
+# @wes-dev/pr
+
+`@wes-dev/pr` is a local-first pull request review workspace. It loads one known PR, groups its changes for a human reviewer, keeps draft state in server memory, and submits a normal GitHub review through the locally authenticated GitHub CLI.
+
+The LLM organizes changes only. It does not generate automated review findings or make approve/reject decisions.
+
+## Run
+
+Run from a checkout of the GitHub repository that owns the PR:
+
+```sh
+npx @wes-dev/pr owner/repo#123
+```
+
+Pull request URLs work too:
+
+```sh
+npx @wes-dev/pr https://github.com/owner/repo/pull/123
+```
+
+Or open the local workspace and enter a PR number in the UI:
+
+```sh
+npx @wes-dev/pr
+```
+
+The CLI binds to `127.0.0.1` by default, tries port `7777` and increments until it finds a free port, prints the URL, and never opens a browser. `--host` and `--port` are available when needed.
+
+The package uses the locally installed [`gh`](https://cli.github.com/) CLI. Authenticate it first with `gh auth login`. The UI has separate Repository and PR number inputs; pasting a GitHub pull request URL into either one fills both fields. A blank repository falls back to the current working directory, and bare `npx @wes-dev/pr 123` remains available for that local-repository case. The tool deliberately does not list or search a large organization’s PRs.
+
+## Review flow
+
+- Load PR metadata, changed files, and patch only when requested.
+- Read current GitHub review comments only when the UI requests them.
+- Keep the loaded PR, groups, draft comments, and review decision in process memory.
+- Add line-specific draft comments, then submit them with a GitHub `COMMENT`, `APPROVE`, or `REQUEST_CHANGES` review.
+
+No database or local PR checkout is created. Restarting the server discards its draft state.
+
+## LLM grouping
+
+The grouping pipeline avoids sending the full patch as one prompt:
+
+1. A manifest pass receives the PR title, description, and file list to propose logical file groups.
+2. One agent per group receives server-side `list_file_chunks` and `read_file_chunk` tools. It can inspect only chunks belonging to its candidate files.
+3. Rejected files are regrouped and retried up to the configured refinement limit.
+4. A final reducer normalizes the resulting groups. The server expands every file group into explicit chunk assignments.
+
+Agents run sequentially by default. Set `maxParallelAgents` when the configured model runtime can handle concurrent requests.
+
+Configuration is read from `~/.config/wes-dev/pr/config.json`; environment variables take precedence.
+
+```json
+{
+  "llm": {
+    "preset": "omlx-openai",
+    "providerName": "omlx",
+    "baseUrl": "http://127.0.0.1:8000/v1",
+    "model": "your-model-name",
+    "apiKey": "replace-me",
+    "headers": {},
+    "queryParams": {},
+    "supportsStructuredOutputs": true,
+    "debug": false,
+    "maxParallelAgents": 1,
+    "maxRefinementRounds": 1,
+    "maxAgentIterations": 8
+  }
+}
+```
+
+`omlx` is treated as an explicit OpenAI-compatible endpoint. For Ollama, use `preset: "ollama-openai"` and set `model`; its local OpenAI-compatible endpoint defaults to `http://127.0.0.1:11434/v1`.
+
+## Development and publishing
+
+```sh
+corepack enable
+pnpm install
+pnpm dev
+```
+
+`pnpm build` creates minified static web assets plus a bundled Node CLI/server. The published package has no runtime dependencies in `package.json`, so `npx` installs the package artifact rather than this project’s source dependency graph.