aicodinggym-cli 0.2.0__tar.gz → 0.4.0__tar.gz

{aicodinggym_cli-0.2.0 → aicodinggym_cli-0.4.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aicodinggym-cli
-Version: 0.2.0
+Version: 0.4.0
 Summary: CLI tool for AI Coding Gym platform
 Author-email: AICodingGym Team <datasmithlab@gmail.com>
 License-Expression: MIT
@@ -20,7 +20,7 @@ Requires-Dist: requests>=2.31.0
 # aicodinggym-cli
 
 CLI tool for the [AI Coding Gym](https://aicodinggym.com) platform.
-Supports two benchmarks: **SWE-bench** (code bug fixes) and **MLE-bench** (ML competitions).
+Supports three benchmarks: **SWE-bench** (code bug fixes), **MLE-bench** (ML competitions), and **Code Review** challenges.
 
 **Install:** `pip install aicodinggym-cli`
 **Entry point:** `aicodinggym`
@@ -43,6 +43,11 @@ aicodinggym swe submit django__django-10097
 aicodinggym mle download spaceship-titanic
 # ... train model, generate predictions ...
 aicodinggym mle submit spaceship-titanic -F predictions.csv
+
+# 4. Code Review: fetch, review, submit
+aicodinggym cr fetch keycloak-0008
+# ... read diff.patch, write your review in review.md ...
+aicodinggym cr submit keycloak-0008 -f review.md
 ```
 
 ---
@@ -154,6 +159,38 @@ aicodinggym mle submit COMPETITION_ID -F FILE [--user-id ID] [--message MSG]
 
 ---
 
+### `aicodinggym cr` — Code Review Commands
+
+#### `aicodinggym cr fetch PROBLEM_ID`
+
+Download the PR diff and create a `review.md` template.
+
+```
+aicodinggym cr fetch PROBLEM_ID [--user-id ID] [--workspace-dir DIR]
+```
+
+Creates in `<workspace>/<problem_id>/`:
+- `diff.patch` — the full diff between base and head branches
+- `review.md` — template to fill in your review (only created if not already present)
+
+#### `aicodinggym cr submit PROBLEM_ID`
+
+Submit your code review.
+
+```
+aicodinggym cr submit PROBLEM_ID -f review.md [--user-id ID]
+aicodinggym cr submit PROBLEM_ID -m "Inline review text"
+echo "My review" | aicodinggym cr submit PROBLEM_ID
+```
+
+| Option | Description |
+|---|---|
+| `-f, --file` | Path to a file containing your review (e.g. `review.md`) |
+| `-m, --message` | Inline review text |
+| stdin | Pipe review text from stdin |
+
+---
+
 ## File Structure
 
 ```

{aicodinggym_cli-0.2.0 → aicodinggym_cli-0.4.0}/README.md

@@ -1,7 +1,7 @@
 # aicodinggym-cli
 
 CLI tool for the [AI Coding Gym](https://aicodinggym.com) platform.
-Supports two benchmarks: **SWE-bench** (code bug fixes) and **MLE-bench** (ML competitions).
+Supports three benchmarks: **SWE-bench** (code bug fixes), **MLE-bench** (ML competitions), and **Code Review** challenges.
 
 **Install:** `pip install aicodinggym-cli`
 **Entry point:** `aicodinggym`
@@ -24,6 +24,11 @@ aicodinggym swe submit django__django-10097
 aicodinggym mle download spaceship-titanic
 # ... train model, generate predictions ...
 aicodinggym mle submit spaceship-titanic -F predictions.csv
+
+# 4. Code Review: fetch, review, submit
+aicodinggym cr fetch keycloak-0008
+# ... read diff.patch, write your review in review.md ...
+aicodinggym cr submit keycloak-0008 -f review.md
 ```
 
 ---
@@ -135,6 +140,38 @@ aicodinggym mle submit COMPETITION_ID -F FILE [--user-id ID] [--message MSG]
 
 ---
 
+### `aicodinggym cr` — Code Review Commands
+
+#### `aicodinggym cr fetch PROBLEM_ID`
+
+Download the PR diff and create a `review.md` template.
+
+```
+aicodinggym cr fetch PROBLEM_ID [--user-id ID] [--workspace-dir DIR]
+```
+
+Creates in `<workspace>/<problem_id>/`:
+- `diff.patch` — the full diff between base and head branches
+- `review.md` — template to fill in your review (only created if not already present)
+
+#### `aicodinggym cr submit PROBLEM_ID`
+
+Submit your code review.
+
+```
+aicodinggym cr submit PROBLEM_ID -f review.md [--user-id ID]
+aicodinggym cr submit PROBLEM_ID -m "Inline review text"
+echo "My review" | aicodinggym cr submit PROBLEM_ID
+```
+
+| Option | Description |
+|---|---|
+| `-f, --file` | Path to a file containing your review (e.g. `review.md`) |
+| `-m, --message` | Inline review text |
+| stdin | Pipe review text from stdin |
+
+---
+
 ## File Structure
 
 ```

{aicodinggym_cli-0.2.0 → aicodinggym_cli-0.4.0}/__init__.py

@@ -1,3 +1,3 @@
 """AI Coding Gym CLI."""
 
-__version__ = "0.2.0"
+__version__ = "0.3.0"

{aicodinggym_cli-0.2.0 → aicodinggym_cli-0.4.0}/aicodinggym_cli.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aicodinggym-cli
-Version: 0.2.0
+Version: 0.4.0
 Summary: CLI tool for AI Coding Gym platform
 Author-email: AICodingGym Team <datasmithlab@gmail.com>
 License-Expression: MIT
@@ -20,7 +20,7 @@ Requires-Dist: requests>=2.31.0
 # aicodinggym-cli
 
 CLI tool for the [AI Coding Gym](https://aicodinggym.com) platform.
-Supports two benchmarks: **SWE-bench** (code bug fixes) and **MLE-bench** (ML competitions).
+Supports three benchmarks: **SWE-bench** (code bug fixes), **MLE-bench** (ML competitions), and **Code Review** challenges.
 
 **Install:** `pip install aicodinggym-cli`
 **Entry point:** `aicodinggym`
@@ -43,6 +43,11 @@ aicodinggym swe submit django__django-10097
 aicodinggym mle download spaceship-titanic
 # ... train model, generate predictions ...
 aicodinggym mle submit spaceship-titanic -F predictions.csv
+
+# 4. Code Review: fetch, review, submit
+aicodinggym cr fetch keycloak-0008
+# ... read diff.patch, write your review in review.md ...
+aicodinggym cr submit keycloak-0008 -f review.md
 ```
 
 ---
@@ -154,6 +159,38 @@ aicodinggym mle submit COMPETITION_ID -F FILE [--user-id ID] [--message MSG]
 
 ---
 
+### `aicodinggym cr` — Code Review Commands
+
+#### `aicodinggym cr fetch PROBLEM_ID`
+
+Download the PR diff and create a `review.md` template.
+
+```
+aicodinggym cr fetch PROBLEM_ID [--user-id ID] [--workspace-dir DIR]
+```
+
+Creates in `<workspace>/<problem_id>/`:
+- `diff.patch` — the full diff between base and head branches
+- `review.md` — template to fill in your review (only created if not already present)
+
+#### `aicodinggym cr submit PROBLEM_ID`
+
+Submit your code review.
+
+```
+aicodinggym cr submit PROBLEM_ID -f review.md [--user-id ID]
+aicodinggym cr submit PROBLEM_ID -m "Inline review text"
+echo "My review" | aicodinggym cr submit PROBLEM_ID
+```
+
+| Option | Description |
+|---|---|
+| `-f, --file` | Path to a file containing your review (e.g. `review.md`) |
+| `-m, --message` | Inline review text |
+| stdin | Pipe review text from stdin |
+
+---
+
 ## File Structure
 
 ```

{aicodinggym_cli-0.2.0 → aicodinggym_cli-0.4.0}/api.py

@@ -1,8 +1,10 @@
 """HTTP API client for the AI Coding Gym backend at aicodinggym.com."""
 
+import os
+
 import requests
 
-API_BASE = "https://aicodinggym.com/api"
+API_BASE = os.environ.get("AICODINGGYM_API_BASE", "https://aicodinggym.com/api")
 TIMEOUT = 30
 
 
@@ -84,6 +86,20 @@ def submit_notification(problem_id: str, user_id: str, commit_hash: str,
     })
 
 
+def fetch_pr(user_id: str, problem_id: str) -> dict:
+    """Fetch CR problem info. Returns {'base_branch': ..., 'head_branch': ..., 'repo_url': ...}."""
+    return _post("code-review-fetch", {"user_id": user_id, "problem_id": problem_id})
+
+
+def cr_submit_review(user_id: str, problem_id: str, review: str) -> dict:
+    """Submit a code review."""
+    return _post("code-review-submit", {
+        "user_id": user_id,
+        "problem_id": problem_id,
+        "review": review,
+    })
+
+
 def mlebench_download_info(user_id: str, competition_id: str, dest_path: str) -> None:
     """Download dataset for an MLE-bench competition directly to dest_path."""
     resp = _get(f"competitions/{competition_id}/download", stream=True)

{aicodinggym_cli-0.2.0 → aicodinggym_cli-0.4.0}/cli.py

@@ -1,7 +1,7 @@
 """AI Coding Gym CLI - main entry point.
 
 A command-line tool for the AI Coding Gym platform (https://aicodinggym.com).
-Supports SWE-bench and MLE-bench challenges.
+Supports SWE-bench, MLE-bench, and Code Review challenges.
 
 SETUP (required before any other command):
     aicodinggym configure --user-id YOUR_USER_ID
@@ -15,6 +15,11 @@ MLE-BENCH WORKFLOW:
     aicodinggym mle download spaceship-titanic
     # ... train model, generate predictions ...
     aicodinggym mle submit spaceship-titanic -F submission.csv
+
+CODE REVIEW WORKFLOW:
+    aicodinggym cr fetch sentry-0001
+    # ... review the diff and write your review ...
+    aicodinggym cr submit sentry-0001 -f review.md
 """
 
 import os
@@ -31,6 +36,8 @@ from . import __version__
 from .api import (
     APIError,
     configure as api_configure,
+    cr_submit_review,
+    fetch_pr as api_fetch_pr,
     fetch_problem as api_fetch_problem,
     mlebench_download_file,
     mlebench_download_info,
@@ -47,11 +54,19 @@ from .git_ops import (
     add_commit_push,
     check_tool_installed,
     clone_repo,
+    clone_repo_cr,
     generate_ssh_key_pair,
     reset_to_setup_commit,
+    run_git_command,
 )
 
 
+def _hyperlink(url: str, text: str | None = None) -> str:
+    """Return an OSC 8 terminal hyperlink. Falls back to plain URL on unsupported terminals."""
+    label = text or url
+    return f"\033]8;;{url}\033\\{label}\033]8;;\033\\"
+
+
 def _error(msg: str) -> None:
     """Print an error message to stderr and exit."""
     click.echo(f"Error: {msg}", err=True)
@@ -228,7 +243,9 @@ def _resolve_key_path(config: dict, creds: dict | None = None) -> Path:
         "  aicodinggym swe fetch django__django-10097\n"
         "  aicodinggym swe submit django__django-10097 --message 'Fix auth bug'\n"
         "  aicodinggym mle download spaceship-titanic\n"
-        "  aicodinggym mle submit spaceship-titanic -F predictions.csv\n\n"
+        "  aicodinggym mle submit spaceship-titanic -F predictions.csv\n"
+        "  aicodinggym cr fetch sentry-0001\n"
+        "  aicodinggym cr submit sentry-0001 -f review.md\n\n"
         "\b\n"
         "WEBSITE:\n"
         "  https://aicodinggym.com\n"
@@ -326,7 +343,7 @@ def configure(user_id: str, workspace_dir: str | None):
             f"  SSH Key:     {private_key_path}\n"
             f"  Config:      ~/.aicodinggym/config.json\n"
             f"\n"
-            f"You can now use 'aicodinggym swe' and 'aicodinggym mle' commands."
+            f"You can now use 'aicodinggym swe', 'aicodinggym mle', and 'aicodinggym cr' commands."
         )
     except APIError as e:
         _error(str(e))
@@ -548,7 +565,7 @@ def swe_submit(problem_id: str, user_id: str | None, message: str | None,
         f"  Branch:  {branch}\n"
         f"  Status:  Pushed and backend notified\n"
         f"\n"
-        f"Your solution has been submitted for evaluation!"
+        f"View results at: {_hyperlink(f'https://aicodinggym.com/challenges/swe/{problem_id}')}"
     )
 
 
@@ -845,6 +862,187 @@ def swe_test(problem_id: str, user_id: str | None, workspace_dir: str | None,
         sys.exit(proc.returncode)
 
 
+# ── cr group ──────────────────────────────────────────────────────────────────
+
+
+@main.group()
+def cr():
+    """Code Review challenges - submit reviews for code diffs.
+
+    \b
+    PREREQUISITE:
+      Run 'aicodinggym configure --user-id YOUR_USER_ID' before using these commands.
+
+    \b
+    WORKFLOW:
+      1. aicodinggym cr fetch CR_PROBLEM_ID             # Download diff + create review.md
+      2. (edit review.md with your findings)
+      3. aicodinggym cr submit CR_PROBLEM_ID -f review.md   # Submit your review
+    """
+    pass
+
+
+@cr.command("fetch")
+@click.argument("problem_id")
+@click.option("--user-id", default=None, help="Override configured user ID.")
+@click.option("--workspace-dir", default=None, type=click.Path(),
+              help="Directory to clone into. Overrides configured workspace.")
+def cr_fetch(problem_id: str, user_id: str | None, workspace_dir: str | None):
+    """Fetch a Code Review problem: downloads the PR diff and creates a review template.
+
+    Clones the repository, generates diff.patch from base→head, and creates
+    review.md as a template to fill in your review.
+
+    \b
+    ARGUMENTS:
+      PROBLEM_ID  The code review problem identifier (e.g., 'keycloak-0008').
+
+    \b
+    EXAMPLE:
+      aicodinggym cr fetch keycloak-0008
+      # Edit review.md in the problem directory, then:
+      aicodinggym cr submit keycloak-0008 -f review.md
+    """
+    config = load_config()
+    uid = _resolve_user_id(config, user_id)
+    workspace = _resolve_workspace(config, workspace_dir)
+
+    try:
+        click.echo(f"Fetching problem '{problem_id}' from server...")
+        data = api_fetch_pr(uid, problem_id)
+    except APIError as e:
+        _error(str(e))
+
+    base_branch = data.get("base_branch")
+    head_branch = data.get("head_branch")
+    repo_url = data.get("repo_url")
+
+    if not (repo_url and repo_url.strip()) or not (base_branch and base_branch.strip()) or not (head_branch and head_branch.strip()):
+        _error("Server did not return required fields (repo_url, base_branch, head_branch).")
+
+    # Save credentials for later submit
+    credentials = load_credentials()
+    credentials[problem_id] = {
+        "repo_url": repo_url,
+        "base_branch": base_branch,
+        "head_branch": head_branch,
+        "user_id": uid,
+        "workspace_dir": str(workspace),
+        "benchmark": "cr",
+    }
+    save_credentials(credentials)
+
+    workspace.mkdir(parents=True, exist_ok=True)
+
+    click.echo(f"Cloning into {workspace / problem_id}...")
+    success, msg = clone_repo_cr(repo_url, base_branch, head_branch,
+                                  problem_id, str(workspace))
+    if not success:
+        _error(msg)
+
+    problem_dir = workspace / problem_id
+
+    # Generate diff.patch
+    diff_result = run_git_command(
+        ["git", "diff", f"{base_branch}..{head_branch}"], str(problem_dir)
+    )
+    diff_path = problem_dir / "diff.patch"
+    diff_path.write_text(diff_result.stdout)
+
+    # Create review.md template if it doesn't exist yet
+    review_path = problem_dir / "review.md"
+    if not review_path.exists():
+        review_path.write_text(
+            f"# Code Review: {problem_id}\n\n"
+            "## Summary\n\n"
+            "<!-- Brief summary of what this PR does -->\n\n"
+            "## Issues Found\n\n"
+            "<!-- List bugs, logic errors, security issues, etc. -->\n\n"
+            "## Suggestions\n\n"
+            "<!-- Optional improvements, style notes, etc. -->\n\n"
+            "## Verdict\n\n"
+            "<!-- Approve / Request Changes / Comment -->\n"
+        )
+
+    cat_cmd = "type" if sys.platform == "win32" else "cat"
+    click.echo(
+        f"\nSuccessfully fetched: {problem_id}\n"
+        f"\n"
+        f"  Diff saved to:     {diff_path}\n"
+        f"  Review template:   {review_path}\n"
+        f"\n"
+        f"Next steps:\n"
+        f"  1. Review the diff:  {cat_cmd} {diff_path}\n"
+        f"  2. Write your review in {review_path}\n"
+        f"  3. Submit:  aicodinggym cr submit {problem_id} -f review.md\n"
+    )
+
+
+@cr.command("submit")
+@click.argument("problem_id")
+@click.option("--user-id", default=None, help="Override configured user ID.")
+@click.option(
+    "-f", "--file", "review_file", type=click.Path(exists=True),
+    help="Path to a file containing your review.",
+)
+@click.option(
+    "-m", "--message", "review_text",
+    help="Inline review text.",
+)
+def cr_submit(problem_id: str, user_id: str | None, review_file: str | None,
+              review_text: str | None):
+    """Submit a code review for a Code Review challenge.
+
+    Reads your review from a file (-f), inline text (-m), or piped stdin,
+    and submits it to the AI Coding Gym server.
+
+    \b
+    ARGUMENTS:
+      PROBLEM_ID  The code review problem identifier (e.g., 'sentry-0001').
+
+    \b
+    EXAMPLE:
+      aicodinggym cr submit sentry-0001 -f review.md
+      aicodinggym cr submit sentry-0001 -m "Found a null pointer bug on line 42"
+      echo "My review" | aicodinggym cr submit sentry-0001
+    """
+    config = load_config()
+    uid = _resolve_user_id(config, user_id)
+
+    # Collect review text (priority: -f > -m > stdin)
+    review = None
+    if review_file:
+        review = Path(review_file).read_text()
+    elif review_text:
+        review = review_text
+    elif not sys.stdin.isatty():
+        review = sys.stdin.read()
+
+    if not review or not review.strip():
+        _error(
+            "No review text provided.\n\n"
+            "Provide your review using one of:\n"
+            "  -f <file>       Read review from a file\n"
+            "  -m \"text\"       Inline review text\n"
+            "  echo ... | ...  Pipe from stdin\n\n"
+            "Example:\n"
+            f"  aicodinggym cr submit {problem_id} -f review.md"
+        )
+
+    try:
+        result = cr_submit_review(uid, problem_id, review.strip())
+    except APIError as e:
+        _error(str(e))
+
+    click.echo(
+        f"\nSuccessfully submitted code review for {problem_id}\n"
+        f"\n"
+        f"  Status:  {result.get('status', 'COMPLETED')}\n"
+        f"\n"
+        f"View results at: {_hyperlink(f'https://aicodinggym.com/challenges/cr/{problem_id}')}"
+    )
+
+
 # ── mle group ────────────────────────────────────────────────────────────────
 
 
@@ -974,4 +1172,4 @@ def mle_submit(competition_id: str, csv_path: str, user_id: str | None,
     )
     if score is not None:
         click.echo(f"  Score:   {score}\n")
-    click.echo("Your prediction has been submitted for scoring!")
+    click.echo(f"View results at: {_hyperlink(f'https://aicodinggym.com/challenges/mle/{competition_id}')}")

{aicodinggym_cli-0.2.0 → aicodinggym_cli-0.4.0}/config.py

@@ -6,6 +6,9 @@ SSH keys are stored in ~/.aicodinggym/{user_id}_id_rsa.
 """
 
 import json
+import os
+import subprocess
+import sys
 from pathlib import Path
 from typing import Any
 
@@ -19,8 +22,22 @@ _CONFIG_FIELDS = ("user_id", "repo_name", "private_key_path", "workspace_dir")
 
 
 def ensure_config_dir() -> Path:
-    """Create the config directory with secure permissions if it doesn't exist."""
-    CONFIG_DIR.mkdir(mode=0o700, exist_ok=True)
+    """Create the config directory with secure permissions if it doesn't exist.
+
+    On Unix/macOS: mode 0o700 (owner-only access).
+    On Windows: removes inherited ACLs and grants full control only to the
+    current user via icacls.
+    """
+    created = not CONFIG_DIR.exists()
+    CONFIG_DIR.mkdir(mode=0o700, parents=True, exist_ok=True)
+    if created and sys.platform == "win32":
+        username = os.environ.get("USERNAME", "")
+        if username:
+            subprocess.run(
+                ["icacls", str(CONFIG_DIR), "/inheritance:r",
+                 "/grant:r", f"{username}:(OI)(CI)(F)"],
+                capture_output=True,
+            )
     return CONFIG_DIR
 
 

aicodinggym_cli-0.4.0/git_ops.py

@@ -0,0 +1,309 @@
+"""Git and SSH key operations for AI Coding Gym CLI."""
+
+import os
+import re
+import shutil
+import subprocess
+import sys
+from pathlib import Path
+from typing import Optional
+
+from .config import ensure_config_dir
+
+
+def _find_git_ssh() -> str | None:
+    """On Windows, find Git for Windows' bundled ssh.exe.
+
+    Windows may have two SSH binaries on PATH: the built-in OpenSSH
+    (C:\\Windows\\System32\\OpenSSH\\ssh.exe) and Git for Windows' MSYS2
+    ssh (C:\\Program Files\\Git\\usr\\bin\\ssh.exe).  System32 is usually
+    first on PATH, so an unqualified 'ssh' resolves to Windows OpenSSH,
+    which can trigger GUI credential dialogs or deadlock when stdout is
+    captured.  This function returns the full path to Git's bundled ssh
+    so we can reference it explicitly in GIT_SSH_COMMAND.
+    """
+    if sys.platform != "win32":
+        return None
+    git_path = shutil.which("git")
+    if not git_path:
+        return None
+    # Walk up from git.exe to find the Git root containing usr/bin/ssh.exe.
+    # Handles cmd/, bin/, and mingw64/bin/ layouts.
+    candidate = Path(git_path).resolve().parent
+    for _ in range(4):
+        ssh = candidate / "usr" / "bin" / "ssh.exe"
+        if ssh.exists():
+            return str(ssh).replace("\\", "/")
+        candidate = candidate.parent
+    return None
+
+
+def _validate_git_ref(name: str, label: str) -> None:
+    """Raise ValueError if name contains suspicious shell metacharacters."""
+    if re.search(r'[;&|`$(){}]', name):
+        raise ValueError(f"Invalid {label}: {name!r}")
+
+
+def _restrict_key_permissions(key_path: Path) -> None:
+    """Restrict an SSH private key file to owner-only access.
+
+    On Unix/macOS: chmod 600 (read/write owner only).
+    On Windows: uses icacls to remove inherited permissions and grant
+    full control only to the current user.  SSH clients on both platforms
+    refuse to use a key whose permissions are too open.
+    """
+    if sys.platform == "win32":
+        # Remove inherited ACLs, then grant only the current user full control.
+        # (F) = Full control, matching chmod 0o600 (owner read+write).
+        key_str = str(key_path)
+        username = os.environ.get("USERNAME", "")
+        if username:
+            subprocess.run(
+                ["icacls", key_str, "/inheritance:r",
+                 "/grant:r", f"{username}:(F)"],
+                capture_output=True,
+            )
+    else:
+        key_path.chmod(0o600)
+
+
+def generate_ssh_key_pair(user_id: str) -> tuple[Path, str]:
+    """Generate an SSH key pair for the user.
+
+    First checks ~/.mcp-keys/ for existing keys matching the user_id.
+    If found, copies them to ~/.aicodinggym/ and reuses them.
+    Otherwise generates new keys in ~/.aicodinggym/{user_id}_id_rsa.
+    Returns (private_key_path, public_key_content).
+    """
+    key_dir = ensure_config_dir()
+    key_path = key_dir / f"{user_id}_id_rsa"
+
+    if not key_path.exists():
+        # Check ~/.mcp-keys/ for existing keys matching user_id
+        mcp_keys_dir = Path.home() / ".mcp-keys"
+        mcp_private = mcp_keys_dir / f"{user_id}_id_rsa"
+        mcp_public = mcp_keys_dir / f"{user_id}_id_rsa.pub"
+
+        if mcp_private.exists() and mcp_public.exists():
+            shutil.copy2(mcp_private, key_path)
+            shutil.copy2(mcp_public, Path(f"{key_path}.pub"))
+            _restrict_key_permissions(key_path)
+        else:
+            if not shutil.which("ssh-keygen"):
+                raise RuntimeError(
+                    "ssh-keygen is not installed or not on PATH.\n"
+                    "On Windows, install Git for Windows (https://git-scm.com) "
+                    "which includes ssh-keygen, or use the OpenSSH optional feature."
+                )
+            result = subprocess.run(
+                ["ssh-keygen", "-t", "rsa", "-b", "4096", "-f", str(key_path),
+                 "-N", "", "-C", f"aicodinggym-{user_id}"],
+                capture_output=True, text=True,
+            )
+            if result.returncode != 0:
+                raise RuntimeError(f"Failed to generate SSH key: {result.stderr}")
+
+    pub_key_path = Path(f"{key_path}.pub")
+    public_key = pub_key_path.read_text().strip()
+    return key_path, public_key
+
+
+def run_git_command(cmd: list[str], cwd: str, key_path: Optional[Path] = None) -> subprocess.CompletedProcess:
+    """Execute a git command with optional SSH key configuration.
+
+    cmd must be a list of arguments (e.g. ["git", "status"]).
+    """
+    env = os.environ.copy()
+    if key_path:
+        # Quote the key path in case it contains spaces (common on Windows).
+        # Use forward slashes — works on all platforms and avoids backslash escaping.
+        quoted_key = str(key_path).replace("\\", "/")
+        # On Windows, use Git for Windows' bundled ssh to avoid Windows native
+        # OpenSSH which can trigger GUI credential dialogs or deadlock when
+        # stdout is captured.  Falls back to bare "ssh" if not found.
+        ssh_bin = _find_git_ssh() or "ssh"
+        # Always use /dev/null for UserKnownHostsFile.  On macOS/Linux this is
+        # the native null device.  On Windows, Git for Windows bundles MSYS2's
+        # ssh which translates /dev/null correctly.  Using os.devnull ("nul")
+        # would break MSYS2's ssh which treats "nul" as a literal filename.
+        # BatchMode=yes prevents any interactive prompts (password, passphrase)
+        # that would cause a hang when stdout/stderr are captured.
+        env["GIT_SSH_COMMAND"] = (
+            f'"{ssh_bin}" -i "{quoted_key}" '
+            f"-o StrictHostKeyChecking=no "
+            f"-o UserKnownHostsFile=/dev/null "
+            f"-o BatchMode=yes"
+        )
+
+    return subprocess.run(cmd, cwd=cwd, capture_output=True, text=True, env=env)
+
+
+def clone_repo(repo_url: str, branch: str, dest_name: str,
+               workspace: str, key_path: Path) -> tuple[bool, str]:
+    """Clone a repo branch into workspace/dest_name.
+
+    Returns (success, message).
+    """
+    problem_dir = Path(workspace) / dest_name
+
+    if problem_dir.exists():
+        # Check if already on the correct branch
+        result = run_git_command(["git", "rev-parse", "--abbrev-ref", "HEAD"], str(problem_dir))
+        if result.returncode == 0 and result.stdout.strip() == branch:
+            pull = run_git_command(["git", "pull", "origin", branch], str(problem_dir), key_path)
+            if pull.returncode != 0:
+                return False, f"Git pull failed:\n{pull.stderr}"
+            return True, f"Already exists. Updated to latest version.\nRepository: {problem_dir}\nBranch: {branch}"
+        return False, (
+            f"Directory {problem_dir} already exists with different content.\n"
+            "Remove it first or use --workspace-dir to specify a different location."
+        )
+
+    cmd = ["git", "clone", "--single-branch", "--branch", branch, "--depth", "1", repo_url, dest_name]
+    result = run_git_command(cmd, workspace, key_path)
+
+    if result.returncode != 0:
+        return False, f"Git clone failed:\n{result.stderr}\nMake sure the branch '{branch}' exists in the repository."
+
+    return True, f"Cloned to: {problem_dir}\nBranch: {branch}"
+
+
+def clone_repo_cr(repo_url: str, base_branch: str, head_branch: str,
+                  dest_name: str, workspace: str,
+                  key_path: Optional[Path] = None) -> tuple[bool, str]:
+    """Clone a code review repo with both base and head branches.
+
+    Clones the base branch first (shallow), then fetches the head branch.
+    Returns (success, message).
+    """
+    _validate_git_ref(base_branch, "base_branch")
+    _validate_git_ref(head_branch, "head_branch")
+    _validate_git_ref(repo_url, "repo_url")
+    _validate_git_ref(dest_name, "dest_name")
+
+    problem_dir = Path(workspace) / dest_name
+
+    if problem_dir.exists():
+        # Already cloned — fetch latest for both branches
+        for branch in (base_branch, head_branch):
+            result = run_git_command(["git", "fetch", "origin", branch], str(problem_dir), key_path)
+            if result.returncode != 0:
+                return False, f"Git fetch failed for {branch}:\n{result.stderr}"
+            result = run_git_command(["git", "branch", "-f", branch, "FETCH_HEAD"], str(problem_dir))
+            if result.returncode != 0:
+                return False, f"Failed to update branch {branch}:\n{result.stderr}"
+        result = run_git_command(["git", "checkout", head_branch], str(problem_dir))
+        if result.returncode != 0:
+            return False, f"Failed to checkout head branch '{head_branch}':\n{result.stderr}"
+        return True, (
+            f"Already exists. Updated both branches.\n"
+            f"Repository: {problem_dir}\n"
+            f"Branches: {base_branch}, {head_branch}"
+        )
+
+    # Clone base branch (shallow); depth 50 needed for diffing between branches
+    cmd = ["git", "clone", "--single-branch", "--branch", base_branch, "--depth", "50", repo_url, dest_name]
+    result = run_git_command(cmd, workspace, key_path)
+    if result.returncode != 0:
+        return False, f"Git clone failed:\n{result.stderr}"
+
+    # Fetch head branch
+    result = run_git_command(["git", "fetch", "origin", head_branch], str(problem_dir), key_path)
+    if result.returncode != 0:
+        return False, f"Failed to fetch head branch '{head_branch}':\n{result.stderr}"
+
+    # Create local head branch tracking the fetched ref
+    result = run_git_command(["git", "branch", "-f", head_branch, "FETCH_HEAD"], str(problem_dir))
+    if result.returncode != 0:
+        return False, f"Failed to create branch {head_branch}:\n{result.stderr}"
+
+    # Check out head branch so the user starts on the code being reviewed
+    result = run_git_command(["git", "checkout", head_branch], str(problem_dir))
+    if result.returncode != 0:
+        return False, f"Failed to checkout head branch '{head_branch}':\n{result.stderr}"
+
+    return True, (
+        f"Cloned to: {problem_dir}\n"
+        f"Branches: {base_branch}, {head_branch}"
+    )
+
+
+def add_commit_push(problem_dir: str, branch: str, key_path: Path,
+                    message: str, force: bool = False) -> tuple[bool, str, str]:
+    """Stage, commit, and push changes.
+
+    Returns (success, message, commit_hash).
+    """
+    pdir = Path(problem_dir)
+
+    # Stage all changes except .github
+    result = run_git_command(["git", "add", "-A", "--", ".", ":(exclude).github"], str(pdir))
+    if result.returncode != 0:
+        return False, f"Git add failed:\n{result.stderr}", ""
+
+    # Check for staged changes
+    status = run_git_command(["git", "diff", "--cached", "--name-only"], str(pdir))
+    if not status.stdout.strip():
+        return False, "No changes to commit. Your working directory is clean.", ""
+
+    # Commit — pass message directly as a list arg; no shell escaping needed
+    result = run_git_command(["git", "commit", "-m", message], str(pdir))
+    if result.returncode != 0:
+        return False, f"Git commit failed:\n{result.stderr}", ""
+
+    # Get commit hash
+    hash_result = run_git_command(["git", "rev-parse", "HEAD"], str(pdir))
+    commit_hash = hash_result.stdout.strip()
+
+    # Push
+    push_cmd = ["git", "push"]
+    if force:
+        push_cmd.append("--force-with-lease")
+    push_cmd += ["origin", branch]
+    result = run_git_command(push_cmd, str(pdir), key_path)
+    if result.returncode != 0:
+        return False, f"Git push failed:\n{result.stderr}", commit_hash
+
+    return True, "Committed and pushed successfully.", commit_hash
+
+
+def reset_to_setup_commit(problem_dir: str) -> tuple[bool, str]:
+    """Reset repo to the original 'Setup SWE-bench instance:' commit.
+
+    Returns (success, message).
+    """
+    log_result = run_git_command(["git", "log", "--format=%H:%s", "--reverse"], problem_dir)
+    if log_result.returncode != 0:
+        return False, f"Git log failed:\n{log_result.stderr}"
+
+    setup_prefix = "Setup SWE-bench instance:"
+    setup_commit = None
+    for line in log_result.stdout.splitlines():
+        parts = line.split(":", 1)
+        if len(parts) != 2:
+            continue
+        commit_hash, subject = parts[0].strip(), parts[1].strip()
+        if subject.startswith(setup_prefix):
+            setup_commit = commit_hash
+            break
+
+    if not setup_commit:
+        return False, (
+            f"Could not find the original setup commit.\n"
+            f"Expected a commit message starting with '{setup_prefix}'."
+        )
+
+    reset = run_git_command(["git", "reset", "--hard", setup_commit], problem_dir)
+    if reset.returncode != 0:
+        return False, f"Git reset failed:\n{reset.stderr}"
+
+    clean = run_git_command(["git", "clean", "-fd"], problem_dir)
+    if clean.returncode != 0:
+        return False, f"Git clean failed:\n{clean.stderr}"
+
+    return True, f"Reset to setup commit {setup_commit[:8]}.\nLocal changes discarded and untracked files removed."
+
+
+def check_tool_installed(tool_name: str) -> bool:
+    """Check if a CLI tool is available on PATH."""
+    return shutil.which(tool_name) is not None

{aicodinggym_cli-0.2.0 → aicodinggym_cli-0.4.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "aicodinggym-cli"
-version = "0.2.0"
+version = "0.4.0"
 description = "CLI tool for AI Coding Gym platform"
 readme = "README.md"
 requires-python = ">=3.10"

aicodinggym_cli-0.2.0/git_ops.py

@@ -1,162 +0,0 @@
-"""Git and SSH key operations for AI Coding Gym CLI."""
-
-import os
-import shutil
-import subprocess
-from pathlib import Path
-from typing import Optional
-
-from .config import ensure_config_dir
-
-
-def generate_ssh_key_pair(user_id: str) -> tuple[Path, str]:
-    """Generate an SSH key pair for the user.
-
-    First checks ~/.mcp-keys/ for existing keys matching the user_id.
-    If found, copies them to ~/.aicodinggym/ and reuses them.
-    Otherwise generates new keys in ~/.aicodinggym/{user_id}_id_rsa.
-    Returns (private_key_path, public_key_content).
-    """
-    key_dir = ensure_config_dir()
-    key_path = key_dir / f"{user_id}_id_rsa"
-
-    if not key_path.exists():
-        # Check ~/.mcp-keys/ for existing keys matching user_id
-        mcp_keys_dir = Path.home() / ".mcp-keys"
-        mcp_private = mcp_keys_dir / f"{user_id}_id_rsa"
-        mcp_public = mcp_keys_dir / f"{user_id}_id_rsa.pub"
-
-        if mcp_private.exists() and mcp_public.exists():
-            shutil.copy2(mcp_private, key_path)
-            shutil.copy2(mcp_public, Path(f"{key_path}.pub"))
-            key_path.chmod(0o600)
-        else:
-            result = subprocess.run(
-                ["ssh-keygen", "-t", "rsa", "-b", "4096", "-f", str(key_path),
-                 "-N", "", "-C", f"aicodinggym-{user_id}"],
-                capture_output=True, text=True,
-            )
-            if result.returncode != 0:
-                raise RuntimeError(f"Failed to generate SSH key: {result.stderr}")
-
-    pub_key_path = Path(f"{key_path}.pub")
-    public_key = pub_key_path.read_text().strip()
-    return key_path, public_key
-
-
-def run_git_command(cmd: str, cwd: str, key_path: Optional[Path] = None) -> subprocess.CompletedProcess:
-    """Execute a git command with optional SSH key configuration."""
-    env = os.environ.copy()
-    if key_path:
-        env["GIT_SSH_COMMAND"] = f"ssh -i {key_path} -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null"
-
-    return subprocess.run(cmd, shell=True, cwd=cwd, capture_output=True, text=True, env=env)
-
-
-def clone_repo(repo_url: str, branch: str, dest_name: str,
-               workspace: str, key_path: Path) -> tuple[bool, str]:
-    """Clone a repo branch into workspace/dest_name.
-
-    Returns (success, message).
-    """
-    problem_dir = Path(workspace) / dest_name
-
-    if problem_dir.exists():
-        # Check if already on the correct branch
-        result = run_git_command("git rev-parse --abbrev-ref HEAD", str(problem_dir))
-        if result.returncode == 0 and result.stdout.strip() == branch:
-            pull = run_git_command(f"git pull origin {branch}", str(problem_dir), key_path)
-            if pull.returncode != 0:
-                return False, f"Git pull failed:\n{pull.stderr}"
-            return True, f"Already exists. Updated to latest version.\nRepository: {problem_dir}\nBranch: {branch}"
-        return False, (
-            f"Directory {problem_dir} already exists with different content.\n"
-            "Remove it first or use --workspace-dir to specify a different location."
-        )
-
-    cmd = f"git clone --single-branch --branch {branch} --depth 1 {repo_url} {dest_name}"
-    result = run_git_command(cmd, workspace, key_path)
-
-    if result.returncode != 0:
-        return False, f"Git clone failed:\n{result.stderr}\nMake sure the branch '{branch}' exists in the repository."
-
-    return True, f"Cloned to: {problem_dir}\nBranch: {branch}"
-
-
-def add_commit_push(problem_dir: str, branch: str, key_path: Path,
-                    message: str, force: bool = False) -> tuple[bool, str, str]:
-    """Stage, commit, and push changes.
-
-    Returns (success, message, commit_hash).
-    """
-    pdir = Path(problem_dir)
-
-    # Stage all changes except .github
-    result = run_git_command("git add -A -- . ':(exclude).github'", str(pdir))
-    if result.returncode != 0:
-        return False, f"Git add failed:\n{result.stderr}", ""
-
-    # Check for staged changes
-    status = run_git_command("git diff --cached --name-only", str(pdir))
-    if not status.stdout.strip():
-        return False, "No changes to commit. Your working directory is clean.", ""
-
-    # Commit
-    safe_msg = message.replace('"', '\\"')
-    result = run_git_command(f'git commit -m "{safe_msg}"', str(pdir))
-    if result.returncode != 0:
-        return False, f"Git commit failed:\n{result.stderr}", ""
-
-    # Get commit hash
-    hash_result = run_git_command("git rev-parse HEAD", str(pdir))
-    commit_hash = hash_result.stdout.strip()
-
-    # Push
-    push_flag = "--force-with-lease " if force else ""
-    result = run_git_command(f"git push {push_flag}origin {branch}", str(pdir), key_path)
-    if result.returncode != 0:
-        return False, f"Git push failed:\n{result.stderr}", commit_hash
-
-    return True, "Committed and pushed successfully.", commit_hash
-
-
-def reset_to_setup_commit(problem_dir: str) -> tuple[bool, str]:
-    """Reset repo to the original 'Setup SWE-bench instance:' commit.
-
-    Returns (success, message).
-    """
-    log_result = run_git_command("git log --format=%H:%s --reverse", problem_dir)
-    if log_result.returncode != 0:
-        return False, f"Git log failed:\n{log_result.stderr}"
-
-    setup_prefix = "Setup SWE-bench instance:"
-    setup_commit = None
-    for line in log_result.stdout.splitlines():
-        parts = line.split(":", 1)
-        if len(parts) != 2:
-            continue
-        commit_hash, subject = parts[0].strip(), parts[1].strip()
-        if subject.startswith(setup_prefix):
-            setup_commit = commit_hash
-            break
-
-    if not setup_commit:
-        return False, (
-            f"Could not find the original setup commit.\n"
-            f"Expected a commit message starting with '{setup_prefix}'."
-        )
-
-    reset = run_git_command(f"git reset --hard {setup_commit}", problem_dir)
-    if reset.returncode != 0:
-        return False, f"Git reset failed:\n{reset.stderr}"
-
-    clean = run_git_command("git clean -fd", problem_dir)
-    if clean.returncode != 0:
-        return False, f"Git clean failed:\n{clean.stderr}"
-
-    return True, f"Reset to setup commit {setup_commit[:8]}.\nLocal changes discarded and untracked files removed."
-
-
-def check_tool_installed(tool_name: str) -> bool:
-    """Check if a CLI tool is available on PATH."""
-    return shutil.which(tool_name) is not None