Executor

1"""Executor for the agent-observability-experiment-analyzer Claude Code skill."""
2
3import os
4import sys
5from typing import Any
6
7_eval_lib_path = os.path.abspath(os.path.join(os.path.dirname(__file__), "../.."))
8if _eval_lib_path not in sys.path:
9    sys.path.insert(0, _eval_lib_path)
10
11from _eval_lib import DEFAULT_TIMEOUT_SECONDS, invoke_skill  # noqa: E402
12from ddeval import BaseProjectExecutor, Config  # noqa: E402
13
14
15class ExperimentAnalyzerExecutor(BaseProjectExecutor):
16    def execute_single(
17        self, input_data: dict[str, Any], config: Config | None
18    ) -> dict[str, Any]:
19        cfg = config.executor_config if config else {}
20        timeout_seconds = cfg.get("timeout_seconds", DEFAULT_TIMEOUT_SECONDS)
21        mcp_env = cfg.get("mcp_env", "prod")
22
23        baseline = input_data.get("baseline")
24        candidate = input_data.get("candidate")
25        experiment_id = input_data.get("experiment_id")
26
27        if not baseline and not candidate and not experiment_id:
28            raise ValueError(
29                "input_data must contain 'baseline', 'candidate', or 'experiment_id'"
30            )
31
32        parts = ["/agent-observability-experiment-analyzer"]
33        if baseline:
34            parts.append(baseline)
35        if candidate:
36            parts.append(candidate)
37        if not baseline and not candidate and experiment_id:
38            parts.append(experiment_id)
39        if input_data.get("question"):
40            parts.append(input_data["question"])
41        parts.append("--output agent")
42        prompt = " ".join(parts)
43
44        # prompt is passed to the Claude Agent SDK, not to a shell — not a command injection risk
45        report = invoke_skill(prompt, timeout_seconds=timeout_seconds, mcp_env=mcp_env)
46        return {"report": report, "word_count": len(report.split()) if report else 0}