devcouncil 0.1.1 → 0.3.0

package/src/devcouncil/execution/shell_session.py

@@ -0,0 +1,231 @@
+"""Guarded shell command session for DevCouncil tasks."""
+
+from __future__ import annotations
+
+import logging
+import os
+import shlex
+import shutil
+import subprocess
+import uuid
+from pathlib import Path
+
+from rich.console import Console
+
+from devcouncil.domain.evidence import CommandResult
+from devcouncil.domain.task import Task
+from devcouncil.execution.checkpoints import CheckpointService
+from devcouncil.execution.hook_policy import HookPolicy
+from devcouncil.storage.db import get_db
+from devcouncil.storage.native import ShellCommandRepository, ShellSessionRepository, TaskLeaseRepository
+from devcouncil.storage.repositories import EvidenceRepository, TaskRepository
+from devcouncil.telemetry.traces import TraceLogger
+
+logger = logging.getLogger(__name__)
+
+
+class ShellBackend:
+    def run_command(self, command: str, cwd: Path, env: dict[str, str] | None = None) -> subprocess.CompletedProcess[str]:
+        raise NotImplementedError
+
+
+class CommandLoopBackend(ShellBackend):
+    def run_command(self, command: str, cwd: Path, env: dict[str, str] | None = None) -> subprocess.CompletedProcess[str]:
+        args = shlex.split(command, posix=(os.name != "nt"))
+        return subprocess.run(
+            args,
+            cwd=cwd,
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+            errors="replace",
+            env=env,
+        )
+
+
+class ShellWrappedBackend(ShellBackend):
+    """Runs each command through an explicit shell (pwsh/bash/zsh/...)."""
+
+    _LAUNCHERS = {
+        "pwsh": ["pwsh", "-NoProfile", "-Command"],
+        "powershell": ["powershell", "-NoProfile", "-Command"],
+        "bash": ["bash", "-lc"],
+        "zsh": ["zsh", "-lc"],
+        "sh": ["sh", "-lc"],
+    }
+
+    def __init__(self, shell: str):
+        launcher = self._LAUNCHERS.get(shell)
+        if launcher is None:
+            supported = ", ".join(sorted(self._LAUNCHERS))
+            raise ValueError(f"Unknown shell backend '{shell}'. Use auto or one of: {supported}.")
+        if not shutil.which(launcher[0]):
+            raise ValueError(f"Shell '{launcher[0]}' is not installed or not on PATH.")
+        self.launcher = launcher
+
+    def run_command(self, command: str, cwd: Path, env: dict[str, str] | None = None) -> subprocess.CompletedProcess[str]:
+        return subprocess.run(
+            [*self.launcher, command],
+            cwd=cwd,
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+            errors="replace",
+            env=env,
+        )
+
+
+_EVIDENCE_COMMAND_HINTS = ("pytest", "ruff", "mypy", "npm test", "npm run lint", "npm run typecheck")
+
+console = Console()
+
+
+class GuardedShellSession:
+    def __init__(self, project_root: Path, task: Task, *, shell: str = "auto"):
+        self.project_root = project_root.resolve()
+        self.task = task
+        self.shell = shell
+        # HookPolicy (not the bare TaskPolicyEngine) so a chained command is split into
+        # its segments and EACH is allowlisted, plus git-safety denies (force-push,
+        # --no-verify, protected-branch reset). Critical for the `--shell bash/zsh`
+        # backend, which hands the whole string to a real shell that interprets
+        # `;`/`&&`/`|` — a single whole-string allowlist match would let an allowed
+        # prefix smuggle an arbitrary command past the gate.
+        self.policy = HookPolicy(self.project_root)
+        self.backend: ShellBackend = (
+            CommandLoopBackend() if shell in {"auto", "loop"} else ShellWrappedBackend(shell)
+        )
+        self.lease_token: str | None = None
+        self.session_id: str | None = None
+        self.log_dir = self.project_root / ".devcouncil" / "logs"
+        self.log_dir.mkdir(parents=True, exist_ok=True)
+
+    def start(self, *, force: bool = False) -> None:
+        db = get_db(self.project_root)
+        if not db:
+            raise RuntimeError("DevCouncil not initialized.")
+        with db.get_session() as session:
+            lease = TaskLeaseRepository(session).acquire(
+                self.task.id,
+                owner="dev shell",
+                agent=self.shell,
+                force=force,
+            )
+            self.lease_token = lease.lease_token
+            shell_session = ShellSessionRepository(session).start(
+                self.task.id,
+                self.shell,
+                str(self.project_root),
+                lease_id=lease.id,
+            )
+            self.session_id = shell_session.id
+            self.task.status = "running"
+            TaskRepository(session).save(self.task)
+        CheckpointService(self.project_root).create_before(self.task.id)
+
+    def finish(self) -> None:
+        db = get_db(self.project_root)
+        if not db or not self.lease_token:
+            return
+        with db.get_session() as session:
+            if self.session_id:
+                ShellSessionRepository(session).finish(self.session_id, "finished")
+            TaskLeaseRepository(session).release(self.task.id, self.lease_token)
+
+    def run_one(self, command: str) -> int:
+        normalized = " ".join(command.split())
+        decision = self.policy.evaluate_command(normalized, self.task)
+        log_id = uuid.uuid4().hex[:8]
+        stdout_path = self.log_dir / f"{self.task.id}-{log_id}.stdout.log"
+        stderr_path = self.log_dir / f"{self.task.id}-{log_id}.stderr.log"
+
+        if decision.action == "deny":
+            logger.warning("Shell command DENIED for %s: %s (%s)", self.task.id, normalized, decision.reason)
+            self._record_command(normalized, "denied", reason=decision.reason)
+            TraceLogger(self.project_root).log_event(
+                "shell_command_denied",
+                {"command": normalized, "reason": decision.reason},
+                task_id=self.task.id,
+            )
+            # Tell the user *why* — a silent non-zero exit is unactionable.
+            console.print(
+                f"[red]Command denied for {self.task.id}:[/red] {decision.reason or 'not permitted by task policy.'}"
+            )
+            console.print(
+                "[dim]Add it to the task's allowed_commands, or run it outside DevCouncil.[/dim]"
+            )
+            return 1
+
+        logger.info("Shell command for %s: %s", self.task.id, normalized)
+        try:
+            result = self.backend.run_command(normalized, self.project_root)
+        except (NotImplementedError, FileNotFoundError, OSError) as exc:
+            logger.warning("Shell command could not run for %s: %s (%s)", self.task.id, normalized, exc)
+            self._record_command(normalized, "denied", reason=str(exc))
+            console.print(f"[red]Could not run '{normalized}':[/red] {exc}")
+            return 1
+
+        stdout_path.write_text(result.stdout or "", encoding="utf-8")
+        stderr_path.write_text(result.stderr or "", encoding="utf-8")
+        # Echo the command output so the guarded shell is actually usable.
+        if result.stdout:
+            console.print(result.stdout, end="", markup=False, highlight=False)
+        if result.stderr:
+            console.print(result.stderr, end="", markup=False, highlight=False, style="dim")
+        status = "finished" if result.returncode == 0 else "failed"
+        self._record_command(
+            normalized,
+            status,
+            exit_code=result.returncode,
+            stdout_path=str(stdout_path),
+            stderr_path=str(stderr_path),
+        )
+        if any(hint in normalized for hint in _EVIDENCE_COMMAND_HINTS):
+            self._save_command_evidence(normalized, result.returncode, result.stdout or result.stderr or "")
+        TraceLogger(self.project_root).log_event(
+            "shell_command_finished",
+            {"command": normalized, "exit_code": result.returncode},
+            task_id=self.task.id,
+        )
+        return result.returncode
+
+    def _record_command(
+        self,
+        command: str,
+        status: str,
+        *,
+        exit_code: int | None = None,
+        reason: str = "",
+        stdout_path: str = "",
+        stderr_path: str = "",
+    ) -> None:
+        db = get_db(self.project_root)
+        if not db:
+            return
+        with db.get_session() as session:
+            ShellCommandRepository(session).record(
+                self.task.id,
+                command,
+                status,
+                session_id=self.session_id,
+                exit_code=exit_code,
+                reason=reason,
+                stdout_path=stdout_path,
+                stderr_path=stderr_path,
+            )
+
+    def _save_command_evidence(self, command: str, exit_code: int, summary: str) -> None:
+        db = get_db(self.project_root)
+        if not db:
+            return
+        with db.get_session() as session:
+            EvidenceRepository(session).save_command_result(
+                self.task.id,
+                CommandResult(
+                    command=command,
+                    exit_code=exit_code,
+                    stdout_path="",
+                    stderr_path="",
+                    summary=summary[:500],
+                ),
+            )

package/src/devcouncil/execution/task_runner.py

@@ -1,11 +1,15 @@
 import hashlib
+import os
 import subprocess
 import logging
 import re
 import shlex
 from pathlib import Path
-from typing import Literal, Optional
+from typing import Literal, Optional, TYPE_CHECKING
 from devcouncil.domain.task import Task
+
+if TYPE_CHECKING:
+    from devcouncil.app.config import DevCouncilConfig
 from devcouncil.execution.permissions import PermissionManager
 from devcouncil.domain.evidence import CommandResult
 from devcouncil.app.errors import ExecutionError
@@ -24,6 +28,16 @@ class TaskRunner:
         self.project_root = project_root
         self.permissions = permission_manager
         self.patch_engine = PatchEngine(project_root)
+        # Load config and build the trace logger once: run_command previously
+        # re-parsed config.yaml on every command, and each operation rebuilt a
+        # TraceLogger. Both are reusable for the lifetime of the runner.
+        self.config: "Optional[DevCouncilConfig]"
+        try:
+            from devcouncil.app.config import load_config
+            self.config = load_config(project_root)
+        except Exception:
+            self.config = None
+        self.tracer = TraceLogger(project_root)
 
     def _validate_path_within_root(self, path: str) -> None:
         """Ensure a path resolves to a location within the project root."""
@@ -36,7 +50,7 @@ class TaskRunner:
             self._validate_path_within_root(path)
             self.permissions.validate_action("file_write", path, task, operation=operation)
         applied = self.patch_engine.apply_patch(patch)
-        TraceLogger(self.project_root).log_event(
+        self.tracer.log_event(
             "tool_patch_applied",
             {"paths": sorted(changes), "success": applied},
             task_id=task.id,
@@ -106,15 +120,16 @@ class TaskRunner:
         logger.info(f"Executing authorized command: {command}")
         
         try:
-            from devcouncil.app.config import load_config
-            config = load_config(self.project_root)
-            timeout = config.execution.command_timeout
+            timeout = self.config.execution.command_timeout if self.config is not None else 300
         except Exception:
             timeout = 300
 
         try:
             result = subprocess.run(
-                shlex.split(command, posix=False),
+                # POSIX-correct tokenization on POSIX hosts (so quoted args like
+                # `pytest -k "a and b"` split right and match the allowlist), Windows
+                # rules on Windows. Matches shell_session.py.
+                shlex.split(command, posix=(os.name != "nt")),
                 shell=False,
                 capture_output=True,
                 text=True,
@@ -129,7 +144,7 @@ class TaskRunner:
             stderr_path = self._save_command_log(task.id, command, "stderr", stderr)
             stdout_summary = redact_string(stdout[-500:])
             stderr_summary = redact_string(stderr[-500:])
-            TraceLogger(self.project_root).log_event(
+            self.tracer.log_event(
                 "command_executed",
                 {"command": command, "exit_code": result.returncode},
                 task_id=task.id,
@@ -149,14 +164,14 @@ class TaskRunner:
         """Write content to a file if allowed by permissions."""
         self._validate_path_within_root(path)
         full_path = resolve_project_path(self.project_root, path)
-        operation = "modify" if full_path.exists() else "create"
+        operation: Literal["create", "modify"] = "modify" if full_path.exists() else "create"
         self.permissions.validate_action("file_write", path, task, operation=operation)
 
         logger.info(f"Writing authorized file: {path}")
         try:
             full_path.parent.mkdir(parents=True, exist_ok=True)
             full_path.write_text(content, encoding="utf-8")
-            TraceLogger(self.project_root).log_event(
+            self.tracer.log_event(
                 "file_written",
                 {"path": path},
                 task_id=task.id,