open-codev-workflow 0.1.1__tar.gz → 0.1.4__tar.gz

{open_codev_workflow-0.1.1/src/open_codev_workflow.egg-info → open_codev_workflow-0.1.4}/PKG-INFO

@@ -1,13 +1,13 @@
 Metadata-Version: 2.4
 Name: open-codev-workflow
-Version: 0.1.1
+Version: 0.1.4
 Summary: Human-guided AI software delivery for real repositories.
 Author: Martin Urban
 License-Expression: BSD-3-Clause
 Project-URL: Homepage, https://github.com/urban233/CoDev
 Project-URL: Repository, https://github.com/urban233/CoDev
 Project-URL: Issues, https://github.com/urban233/CoDev/issues
-Keywords: ai,developer-tools,software-delivery,codex,opencode
+Keywords: ai,developer-tools,software-delivery,antigravity,codex,junie,opencode
 Classifier: Development Status :: 3 - Alpha
 Classifier: Environment :: Console
 Classifier: Intended Audience :: Developers
@@ -106,23 +106,122 @@ codev remove --target ../my-project --dry-run
 `init`, `diff`, and `update` preflight the entire operation. A locally modified
 managed file becomes a visible conflict; CoDev never silently replaces it.
 
+## GitHub Pull Request reviews
+
+The installed `pr-review` skill reviews an existing GitHub Pull Request and can
+prepare validated inline comments for the exact PR head. It uses the GitHub CLI
+credential store by default, so agents do not need to read or print a token.
+
+### Install and authenticate GitHub CLI on Windows
+
+Install the official package from PowerShell with WinGet:
+
+```powershell
+winget install --id GitHub.cli --source winget
+```
+
+Open a new Windows Terminal window after installation, then verify and sign in:
+
+```powershell
+gh --version
+gh auth login --web
+gh auth status --active
+```
+
+Choose `GitHub.com`, the HTTPS protocol, and the browser login flow. `gh` stores
+the credential using the Windows credential store when available. See the
+[official Windows installation guide](https://github.com/cli/cli/blob/trunk/docs/install_windows.md)
+and [`gh auth login` documentation](https://cli.github.com/manual/gh_auth_login).
+
+Run the PR publisher in dry-run mode first:
+
+```powershell
+python .agents\skills\pr-review\scripts\publish_review.py `
+  --repo OWNER/REPO `
+  --pr 123 `
+  --review review.json
+```
+
+The publisher automatically uses authenticated `gh api` when no
+`GITHUB_TOKEN` or `GH_TOKEN` is set. Use `--auth gh` to require that backend or
+`--auth token` for headless environments that provide a token variable. Add
+`--publish` only after explicitly authorizing a GitHub review, and use
+`--submit comment` only when it should be submitted immediately.
+
+If a desktop agent does not inherit the Windows machine PATH, the publisher
+also checks the standard `C:\\Program Files\\GitHub CLI\\gh.exe` location. For a
+custom installation, set `CODEV_GH_PATH` to the full path of `gh.exe`.
+
+To copy the already-authenticated `gh` credential into `GH_TOKEN` for the
+current PowerShell process and the CLI started from it, dot-source the bundled
+helper:
+
+```powershell
+. .agents\\skills\\pr-review\\scripts\\set-github-token.ps1
+```
+
+The helper calls `gh auth token` without printing the result and does not
+persist it. The `gh` credential must be valid in the same process context. This
+is useful for a CLI that needs `GH_TOKEN`; launch that CLI from the shell where
+the helper has been dot-sourced. Do not put the token in a repository file or
+command-line argument.
+
+One-line equivalent:
+
+```powershell
+$g=Get-Command gh -ErrorAction SilentlyContinue;if($g){$p=$g.Source}else{$p='C:\\Program Files\\GitHub CLI\\gh.exe'};$env:GH_TOKEN=(& $p auth token --hostname github.com 2>$null).Trim()
+```
+
+Fetch the complete GitHub PR context before asking an agent to review it:
+
+```powershell
+python .agents\\skills\\pr-review\\scripts\\publish_review.py `
+  --repo OWNER/REPO `
+  --pr 123 `
+  --fetch `
+  --output-dir .codev\\pr-review\\123
+```
+
+This writes PR metadata, the patch, changed files, commits, reviews, comments,
+and check runs. Use repeated `--include metadata`, `--include diff`, or other
+parts to fetch a smaller set.
+
+The installed Junie project command is also available directly inside Junie:
+
+```text
+/pr-review repo=OWNER/REPO pr=123
+```
+
+Project-specific Junie commands live under `.junie/commands`, so this command
+is versioned with the repository and appears in Junie’s `/` command list.
+
 ## What gets installed
 
 ```text
 my-project/
 ├── AGENTS.md                         # a managed policy block; local text survives
-├── .agents/skills/                   # seven lifecycle skills
+├── .agents/skills/                   # lifecycle, PR, and specialist review skills
+├── .agents/agents/                   # Antigravity subagents
 ├── .opencode/agents/                 # orchestrator, builder, reviewer
 ├── .opencode/opencode.json           # safely merged; existing agent settings survive
+├── .junie/agents/                    # Junie subagents
 ├── docs/                             # workflow, prompts, handbooks, cookbook
 ├── evals/development-workflow/       # behavioral scenarios
 ├── scripts/                          # deterministic validators
 └── .codev/lock.json                # installed version and source hashes
 ```
 
-Use `--platform codex` to omit the OpenCode adapter. Use `--platform opencode`
-or `--platform all` for the three-agent OpenCode topology. Core skills and
-human/AI workflow references are installed for every platform.
+Use `--platform codex` to omit the OpenCode, Junie, and Antigravity adapters.
+Use `--platform opencode`, `--platform junie`, or `--platform antigravity` to
+select one adapter, or use `--platform all` for every supported platform. Core
+skills and human/AI workflow references are installed for every platform.
+
+To add an adapter to an existing installation, pass it to `update`, for example
+`codev update --platform junie`. Use `diff --platform junie` to preview the
+platform expansion first.
+
+The Antigravity adapter follows its official workspace location:
+`.agents/agents/<name>.md`.
 
 ## Design principles
 

{open_codev_workflow-0.1.1 → open_codev_workflow-0.1.4}/README.md

@@ -77,23 +77,122 @@ codev remove --target ../my-project --dry-run
 `init`, `diff`, and `update` preflight the entire operation. A locally modified
 managed file becomes a visible conflict; CoDev never silently replaces it.
 
+## GitHub Pull Request reviews
+
+The installed `pr-review` skill reviews an existing GitHub Pull Request and can
+prepare validated inline comments for the exact PR head. It uses the GitHub CLI
+credential store by default, so agents do not need to read or print a token.
+
+### Install and authenticate GitHub CLI on Windows
+
+Install the official package from PowerShell with WinGet:
+
+```powershell
+winget install --id GitHub.cli --source winget
+```
+
+Open a new Windows Terminal window after installation, then verify and sign in:
+
+```powershell
+gh --version
+gh auth login --web
+gh auth status --active
+```
+
+Choose `GitHub.com`, the HTTPS protocol, and the browser login flow. `gh` stores
+the credential using the Windows credential store when available. See the
+[official Windows installation guide](https://github.com/cli/cli/blob/trunk/docs/install_windows.md)
+and [`gh auth login` documentation](https://cli.github.com/manual/gh_auth_login).
+
+Run the PR publisher in dry-run mode first:
+
+```powershell
+python .agents\skills\pr-review\scripts\publish_review.py `
+  --repo OWNER/REPO `
+  --pr 123 `
+  --review review.json
+```
+
+The publisher automatically uses authenticated `gh api` when no
+`GITHUB_TOKEN` or `GH_TOKEN` is set. Use `--auth gh` to require that backend or
+`--auth token` for headless environments that provide a token variable. Add
+`--publish` only after explicitly authorizing a GitHub review, and use
+`--submit comment` only when it should be submitted immediately.
+
+If a desktop agent does not inherit the Windows machine PATH, the publisher
+also checks the standard `C:\\Program Files\\GitHub CLI\\gh.exe` location. For a
+custom installation, set `CODEV_GH_PATH` to the full path of `gh.exe`.
+
+To copy the already-authenticated `gh` credential into `GH_TOKEN` for the
+current PowerShell process and the CLI started from it, dot-source the bundled
+helper:
+
+```powershell
+. .agents\\skills\\pr-review\\scripts\\set-github-token.ps1
+```
+
+The helper calls `gh auth token` without printing the result and does not
+persist it. The `gh` credential must be valid in the same process context. This
+is useful for a CLI that needs `GH_TOKEN`; launch that CLI from the shell where
+the helper has been dot-sourced. Do not put the token in a repository file or
+command-line argument.
+
+One-line equivalent:
+
+```powershell
+$g=Get-Command gh -ErrorAction SilentlyContinue;if($g){$p=$g.Source}else{$p='C:\\Program Files\\GitHub CLI\\gh.exe'};$env:GH_TOKEN=(& $p auth token --hostname github.com 2>$null).Trim()
+```
+
+Fetch the complete GitHub PR context before asking an agent to review it:
+
+```powershell
+python .agents\\skills\\pr-review\\scripts\\publish_review.py `
+  --repo OWNER/REPO `
+  --pr 123 `
+  --fetch `
+  --output-dir .codev\\pr-review\\123
+```
+
+This writes PR metadata, the patch, changed files, commits, reviews, comments,
+and check runs. Use repeated `--include metadata`, `--include diff`, or other
+parts to fetch a smaller set.
+
+The installed Junie project command is also available directly inside Junie:
+
+```text
+/pr-review repo=OWNER/REPO pr=123
+```
+
+Project-specific Junie commands live under `.junie/commands`, so this command
+is versioned with the repository and appears in Junie’s `/` command list.
+
 ## What gets installed
 
 ```text
 my-project/
 ├── AGENTS.md                         # a managed policy block; local text survives
-├── .agents/skills/                   # seven lifecycle skills
+├── .agents/skills/                   # lifecycle, PR, and specialist review skills
+├── .agents/agents/                   # Antigravity subagents
 ├── .opencode/agents/                 # orchestrator, builder, reviewer
 ├── .opencode/opencode.json           # safely merged; existing agent settings survive
+├── .junie/agents/                    # Junie subagents
 ├── docs/                             # workflow, prompts, handbooks, cookbook
 ├── evals/development-workflow/       # behavioral scenarios
 ├── scripts/                          # deterministic validators
 └── .codev/lock.json                # installed version and source hashes
 ```
 
-Use `--platform codex` to omit the OpenCode adapter. Use `--platform opencode`
-or `--platform all` for the three-agent OpenCode topology. Core skills and
-human/AI workflow references are installed for every platform.
+Use `--platform codex` to omit the OpenCode, Junie, and Antigravity adapters.
+Use `--platform opencode`, `--platform junie`, or `--platform antigravity` to
+select one adapter, or use `--platform all` for every supported platform. Core
+skills and human/AI workflow references are installed for every platform.
+
+To add an adapter to an existing installation, pass it to `update`, for example
+`codev update --platform junie`. Use `diff --platform junie` to preview the
+platform expansion first.
+
+The Antigravity adapter follows its official workspace location:
+`.agents/agents/<name>.md`.
 
 ## Design principles
 

{open_codev_workflow-0.1.1 → open_codev_workflow-0.1.4}/pyproject.toml

@@ -4,13 +4,13 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "open-codev-workflow"
-version = "0.1.1"
+version = "0.1.4"
 description = "Human-guided AI software delivery for real repositories."
 readme = "README.md"
 requires-python = ">=3.11"
 license = "BSD-3-Clause"
 authors = [{ name = "Martin Urban" }]
-keywords = ["ai", "developer-tools", "software-delivery", "codex", "opencode"]
+keywords = ["ai", "developer-tools", "software-delivery", "antigravity", "codex", "junie", "opencode"]
 classifiers = [
   "Development Status :: 3 - Alpha",
   "Environment :: Console",
@@ -55,7 +55,10 @@ codev_workflow = [
   "bundle/.agents/skills/*/assets/*",
   "bundle/.agents/skills/*/references/*",
   "bundle/.agents/skills/*/scripts/*",
+  "bundle/.agents/agents/*.md",
   "bundle/.opencode/agents/*.md",
+  "bundle/.junie/agents/*.md",
+  "bundle/.junie/commands/*.md",
   "bundle/docs/*.md",
   "bundle/docs/for-ai/*.md",
   "bundle/docs/handbooks/*.md",

{open_codev_workflow-0.1.1 → open_codev_workflow-0.1.4}/src/codev_workflow/__init__.py

@@ -2,4 +2,4 @@
 
 __all__ = ["__version__"]
 
-__version__ = "0.1.1"
+__version__ = "0.1.4"

open_codev_workflow-0.1.4/src/codev_workflow/bundle/.agents/agents/builder.md

@@ -0,0 +1,50 @@
+---
+name: builder
+description: Bounded implementation subagent that executes one accepted work-item plan.
+mainAgent: false
+subagent: true
+model: inherit
+commandExecutionPolicy: sandbox
+tools:
+  - view_file
+  - grep_search
+  - replace_file_content
+  - run_command
+---
+
+Implement exactly one bounded work item delegated by the orchestrator. Follow
+`AGENTS.md`, `docs/for-ai/WORKFLOW-AGENTS.md`, and `build-change`. Treat the
+accepted implementation plan and its cited brief/specification/design/API as
+authority; do not redesign them to make coding easier.
+
+Before editing:
+
+1. inspect the actual files, symbols, tests, build commands, conventions, Git
+   state, and nearby comparable implementation;
+2. verify the base snapshot and check for unrelated or concurrent changes;
+3. compare repository facts with the accepted plan; and
+4. return `BLOCKED` with exact evidence when a material mismatch, missing
+   decision, scope expansion, collision, or unavailable validation prevents safe
+   execution. Do not invent the missing answer.
+
+When ready, implement the smallest coherent review purpose. Stay within allowed
+scope, reuse repository patterns, put tests with behavior, and avoid unrelated
+cleanup. Never weaken tests, silently change contracts, add unaccepted
+dependencies, or modify accepted planning artifacts. Keep the repository
+buildable and stop after two failed attempts with the same root cause.
+
+Run the specified formatter, static checks, affected tests, and proportionate
+broader validation. Inspect the complete diff. Return only observable handoff
+information:
+
+- **Delivered:** outcome and behavior;
+- **Changed:** files and components;
+- **Base/head snapshot:** exact values;
+- **Validation actually run:** commands and outcomes;
+- **Acceptance evidence:** criterion mapped to evidence;
+- **Scope deviations:** none or explicitly accepted deviations;
+- **Known limitations:** risks and follow-up; and
+- **Review state:** `AWAITING INDEPENDENT REVIEW`.
+
+Do not invoke another agent, approve the change, commit, push, merge, publish,
+deploy, migrate data, or expand rollout.

open_codev_workflow-0.1.4/src/codev_workflow/bundle/.agents/agents/orchestrator.md

@@ -0,0 +1,62 @@
+---
+name: orchestrator
+description: Human-controlled workflow orchestrator for planning, delegated building, and independent review.
+mainAgent: true
+subagent: true
+model: inherit
+commandExecutionPolicy: sandbox
+---
+
+Act as the human's primary engineering partner. Follow `AGENTS.md`,
+`docs/for-ai/WORKFLOW-AGENTS.md`, and the applicable repository skills. Present
+the work as `Understand`, `Build`, `Review`, or `Ship` and select the lightest
+safe path without requiring the human to know skill names.
+
+For Understand and Ship work, use the applicable lifecycle skill directly.
+Create or revise planning artifacts only when the selected skill requires them
+and the human has authorized the write. Never implement product code while
+acting as orchestrator.
+
+## Three-agent Build protocol
+
+For one ready work item:
+
+1. Read the work item, upstream brief/specification/design/API authority,
+   repository instructions, current code and tests, ownership, and Git state.
+2. Confirm the item is ready. Return unresolved product questions to
+   `define-product`, architectural or contract questions to `design-solution`,
+   and dependency or assignment problems to `plan-delivery`.
+3. Use `build-change` to frame and ground the change. Present the focus card.
+   For delegated, multi-session, cross-component, normal-risk, or higher-risk
+   work, render the complete
+   `.agents/skills/build-change/assets/implementation-plan.template.md` in the
+   conversation. Do not ask the human to write it.
+4. Obtain one precise human decision for any material product, API, data,
+   dependency, architecture, security, destructive, scope, or risk choice. Ask
+   for approval before starting delegated implementation. Do not delegate an
+   unresolved or unaccepted plan.
+5. Use Antigravity's `invoke_subagent` capability to start `builder` with the
+   accepted work item and implementation plan, exact authority links, base
+   commit, allowed scope, integration constraints, validation, and stop
+   conditions. Pass task-local artifacts, not private reasoning or a broad
+   conversation transcript.
+6. When the builder returns, verify that its evidence receipt identifies an
+   exact head snapshot, actual validation, deviations, limitations, and changed
+   files. If evidence is missing, return the task for evidence rather than
+   guessing.
+7. Use `invoke_subagent` to start `reviewer` in a fresh task with the exact
+   base-to-head snapshot, work item, accepted plan, upstream authority, and
+   builder evidence receipt.
+8. If the reviewer returns actionable findings, send the findings and original
+   accepted plan back to `builder`. Do not let the reviewer edit. Repeat review
+   after correction. After two correction attempts with the same root cause,
+   or whenever the accepted plan must change materially, stop for the human
+   with evidence and a recommendation.
+9. Return the final evidence receipt, reviewer decision, residual risks, and
+   exact snapshot. Never claim approval and stop before commit, merge, publish,
+   deploy, migration, or rollout expansion unless the human explicitly grants
+   the corresponding authority.
+
+Keep progress visible at plan acceptance, builder completion, reviewer result,
+and any stop condition. Do not spawn unrelated agents or parallel builders in
+the same worktree.

open_codev_workflow-0.1.4/src/codev_workflow/bundle/.agents/agents/reviewer.md

@@ -0,0 +1,34 @@
+---
+name: reviewer
+description: Independent reviewer for one exact code change.
+mainAgent: false
+subagent: true
+model: inherit
+commandExecutionPolicy: sandbox
+tools:
+  - view_file
+  - grep_search
+  - run_command
+---
+
+Use the `review-change` skill. Review the exact supplied base-to-head diff,
+acceptance criteria, relevant design/API, repository context, and validation
+evidence without relying on the implementing agent's private reasoning.
+
+Confirm the exact base and head snapshots before reviewing. If the diff,
+authority, acceptance criteria, or builder evidence is missing or ambiguous,
+return `BLOCKED BY MISSING EVIDENCE` rather than reconstructing it from chat.
+
+Prioritize correctness, security/privacy, data loss, concurrency, compatibility,
+error behavior, test quality, architecture, scope, maintainability, and rollout.
+Assess tests by whether a small, representative suite catches realistic
+regressions and important boundary behavior; coverage percentages are
+diagnostic only. Do not persist on theoretical, rare, low-impact edge cases
+unless they affect safety, data integrity, compatibility, or likely regressions.
+Lead with actionable findings ordered P0 through P3. Give a tight location,
+evidence, impact, and testable correction for each finding.
+
+Do not edit code or planning artifacts. Do not invent requirements, block on
+personal style, communicate with the builder directly, or authorize merge. End
+with `READY FOR HUMAN APPROVAL`, `CHANGES REQUIRED`, or
+`BLOCKED BY MISSING EVIDENCE` and name residual risks.