cleveragents-core/robot/helper_skill_inline.py

"""Robot Framework helper for inline tool executor smoke tests.

Provides a CLI-style interface for Robot to invoke inline tool
execution, validation, and write-guard checks.
Exit code 0 = success, 1 = failure.

Usage:
    python robot/helper_skill_inline.py execute-simple
    python robot/helper_skill_inline.py execute-timeout
    python robot/helper_skill_inline.py execute-truncate
    python robot/helper_skill_inline.py validate-no-code
    python robot/helper_skill_inline.py write-guard
"""

from __future__ import annotations

import sys
from pathlib import Path

# Ensure the src directory is on the import path.
_SRC = str(Path(__file__).resolve().parents[1] / "src")
if _SRC not in sys.path:
    sys.path.insert(0, _SRC)

from cleveragents.domain.models.core.skill import (  # noqa: E402
    SkillInlineTool,
)
from cleveragents.domain.models.core.tool import (  # noqa: E402
    ToolCapability,
    ToolSource,
)
from cleveragents.skills.context import SkillContext  # noqa: E402
from cleveragents.skills.inline_executor import InlineToolExecutor  # noqa: E402


def _cmd_execute_simple() -> int:
    """Test simple inline tool execution."""
    tool = SkillInlineTool(
        description="Simple print",
        source=ToolSource.CUSTOM,
        code='print("hello-robot")',
    )
    ctx = SkillContext(
        plan_id="plan-robot-inline",
        project_id="proj-robot-inline",
        sandbox_path=Path("/tmp/robot-inline-sandbox"),
    )
    executor = InlineToolExecutor()
    result = executor.execute(tool, ctx, {})
    if not result.success:
        print(f"inline-fail: {result.error_message}")
        return 1
    if "hello-robot" not in str(result.output):
        print(f"inline-fail: unexpected output {result.output}")
        return 1
    print("inline-execute-ok")
    return 0


def _cmd_execute_timeout() -> int:
    """Test inline tool timeout."""
    tool = SkillInlineTool(
        description="Sleeping tool",
        source=ToolSource.CUSTOM,
        code="import time; time.sleep(5)",
    )
    ctx = SkillContext(
        plan_id="plan-robot-timeout",
        project_id="proj-robot-timeout",
        sandbox_path=Path("/tmp/robot-timeout-sandbox"),
    )
    executor = InlineToolExecutor(max_runtime_seconds=0.1)
    result = executor.execute(tool, ctx, {})
    if result.success:
        print("inline-fail: expected timeout failure")
        return 1
    if "timed out" not in (result.error_message or ""):
        print(f"inline-fail: unexpected error: {result.error_message}")
        return 1
    print("inline-timeout-ok")
    return 0


def _cmd_execute_truncate() -> int:
    """Test inline tool output truncation."""
    tool = SkillInlineTool(
        description="Large output tool",
        source=ToolSource.CUSTOM,
        code='print("X" * 2000)',
    )
    ctx = SkillContext(
        plan_id="plan-robot-truncate",
        project_id="proj-robot-truncate",
        sandbox_path=Path("/tmp/robot-truncate-sandbox"),
    )
    executor = InlineToolExecutor(max_output_bytes=100)
    result = executor.execute(tool, ctx, {})
    if not result.success:
        print(f"inline-fail: {result.error_message}")
        return 1
    if not result.truncated:
        print("inline-fail: expected truncated=True")
        return 1
    print("inline-truncate-ok")
    return 0


def _cmd_validate_no_code() -> int:
    """Test validation of tool without code."""
    tool = SkillInlineTool(
        description="No-code tool",
        source=ToolSource.CUSTOM,
        code=None,
    )
    executor = InlineToolExecutor()
    errors = executor.validate_tool(tool)
    if not errors:
        print("inline-fail: expected validation errors")
        return 1
    if not any("non-empty code" in e for e in errors):
        print(f"inline-fail: unexpected errors: {errors}")
        return 1
    print("inline-validate-ok")
    return 0


def _cmd_write_guard() -> int:
    """Test write guard blocks execution in read-only context."""
    tool = SkillInlineTool(
        description="Write tool",
        source=ToolSource.CUSTOM,
        code='print("writing")',
        capability=ToolCapability(writes=True),
    )
    ctx = SkillContext(
        plan_id="plan-robot-guard",
        project_id="proj-robot-guard",
        sandbox_path=Path("/tmp/robot-guard-sandbox"),
        read_only=True,
    )
    executor = InlineToolExecutor()
    result = executor.execute(tool, ctx, {})
    if result.success:
        print("inline-fail: expected write guard to block")
        return 1
    if "denied" not in (result.error_message or "").lower():
        print(f"inline-fail: unexpected error: {result.error_message}")
        return 1
    print("inline-guard-ok")
    return 0


_COMMANDS = {
    "execute-simple": _cmd_execute_simple,
    "execute-timeout": _cmd_execute_timeout,
    "execute-truncate": _cmd_execute_truncate,
    "validate-no-code": _cmd_validate_no_code,
    "write-guard": _cmd_write_guard,
}


def main() -> int:
    """Entry point called by Robot Framework ``Run Process``."""
    if len(sys.argv) < 2:
        print(
            "Usage: helper_skill_inline.py "
            "<execute-simple|execute-timeout|execute-truncate"
            "|validate-no-code|write-guard>"
        )
        return 1

    command = sys.argv[1]
    handler = _COMMANDS.get(command)
    if handler is None:
        print(f"Unknown command: {command}")
        return 1

    return handler()


if __name__ == "__main__":
    sys.exit(main())