feat(scripts): added unified edit benchmark runner with variant selection

- Replaced the standalone chunk and vim benchmark scripts with a unified `scripts/edit-benchmark.py` entrypoint.
- Added `--variant` and `PI_EDIT_VARIANT` handling with validation for supported edit modes.
- Generated benchmark specs dynamically per variant with mode-specific prompts and retry instructions.

scripts/chunk-edit-benchmark.py

@@ -1,36 +0,0 @@
-#!/usr/bin/env python3
-"""
-Chunk edit benchmark: tests chunk-mode edit tool usage across models with a simple edit task.
-"""
-from __future__ import annotations
-
-from edit_benchmark_common import BenchmarkSpec, EDIT_DIFF, EXPECTED_CONTENT, run_benchmark_main
-
-EDIT_PROMPT = f"""\
-Use the `read` tool to inspect `test.rs`, then use the `edit` tool in chunk mode to make `test.rs` exactly match the requested change.
-
-Apply this diff:
-```diff
-{EDIT_DIFF}```
-
-Final expected file content:
-```python
-{EXPECTED_CONTENT}```
-"""
-
-CHUNK_BENCHMARK = BenchmarkSpec(
-    description="Benchmark chunk-mode edit tool across models with simple edit tasks.",
-    workspace_prefix="chunk-benchmark",
-    tools=("edit", "read"),
-    env={"PI_EDIT_VARIANT": "chunk", "PI_STRICT_EDIT_MODE": "1"},
-    initial_prompt=EDIT_PROMPT,
-    retry_instruction='Use `read(path="test.rs")` to refresh chunk selectors if needed, then try again using the edit tool.',
-)
-
-
-def main() -> int:
-    return run_benchmark_main(CHUNK_BENCHMARK)
-
-
-if __name__ == "__main__":
-    raise SystemExit(main())

scripts/edit-benchmark.py

@@ -0,0 +1,90 @@
+#!/usr/bin/env python3
+"""
+Edit benchmark: tests the edit tool across models with a simple edit task.
+
+Select the edit variant via the PI_EDIT_VARIANT env var (e.g. `chunk`, `vim`,
+`hashline`, `replace`, `patch`, `apply_patch`) or `--variant`.
+
+Examples:
+	PI_EDIT_VARIANT=chunk scripts/edit-benchmark.py
+	PI_EDIT_VARIANT=vim   scripts/edit-benchmark.py
+	scripts/edit-benchmark.py --variant hashline
+"""
+from __future__ import annotations
+
+import os
+import sys
+
+from edit_benchmark_common import BenchmarkSpec, EDIT_DIFF, EXPECTED_CONTENT, run_benchmark_main
+
+# Variants recognised by packages/coding-agent/src/utils/edit-mode.ts.
+VALID_VARIANTS = ("replace", "patch", "hashline", "chunk", "vim", "apply_patch")
+
+
+def _extract_variant_arg() -> str | None:
+	"""Pop `--variant <value>` (or `--variant=<value>`) from sys.argv before argparse in common runs."""
+	argv = sys.argv
+	for i, arg in enumerate(argv[1:], start=1):
+		if arg == "--variant" and i + 1 < len(argv):
+			value = argv[i + 1]
+			del argv[i : i + 2]
+			return value
+		if arg.startswith("--variant="):
+			value = arg.split("=", 1)[1]
+			del argv[i]
+			return value
+	return None
+
+
+def _resolve_variant() -> str:
+	cli_variant = _extract_variant_arg()
+	variant = cli_variant or os.environ.get("PI_EDIT_VARIANT")
+	if not variant:
+		raise SystemExit(
+			"edit-benchmark: set PI_EDIT_VARIANT=<variant> or pass --variant <variant>.\n"
+			f"Valid variants: {', '.join(VALID_VARIANTS)}"
+		)
+	if variant not in VALID_VARIANTS:
+		raise SystemExit(
+			f"edit-benchmark: unknown variant '{variant}'.\n"
+			f"Valid variants: {', '.join(VALID_VARIANTS)}"
+		)
+	return variant
+
+
+def build_spec(variant: str) -> BenchmarkSpec:
+	mode_phrase = f"in {variant} mode"
+	prompt = (
+		f"Use the `read` tool to inspect `test.rs`, then use the `edit` tool {mode_phrase} "
+		f"to make `test.rs` exactly match the requested change.\n"
+		f"\n"
+		f"Apply this diff:\n"
+		f"```diff\n"
+		f"{EDIT_DIFF}```\n"
+		f"\n"
+		f"Final expected file content:\n"
+		f"```rust\n"
+		f"{EXPECTED_CONTENT}```\n"
+	)
+	retry = (
+		'Use `read(path="test.rs")` to refresh chunk selectors if needed, then try again using the edit tool.'
+		if variant == "chunk"
+		else f"Please try again using the edit tool {mode_phrase}."
+	)
+	return BenchmarkSpec(
+		description=f"Benchmark edit tool in {variant} mode across models with simple edit tasks.",
+		workspace_prefix=f"{variant}-benchmark",
+		tools=("edit", "read"),
+		env={"PI_EDIT_VARIANT": variant, "PI_STRICT_EDIT_MODE": "1"},
+		initial_prompt=prompt,
+		retry_instruction=retry,
+	)
+
+
+def main() -> int:
+	variant = _resolve_variant()
+	return run_benchmark_main(build_spec(variant))
+
+
+if __name__ == "__main__":
+	raise SystemExit(main())

scripts/vim-edit-benchmark.py

@@ -1,37 +0,0 @@
-#!/usr/bin/env python3
-"""
-Vim edit-mode benchmark: tests the edit tool in vim mode across models with a simple edit task.
-"""
-from __future__ import annotations
-
-from edit_benchmark_common import BenchmarkSpec, EDIT_DIFF, EXPECTED_CONTENT, run_benchmark_main
-
-
-EDIT_PROMPT = f"""\
-Use the `read` tool to inspect `test.rs`, then use the `edit` tool in vim mode to make `test.rs` exactly match the requested change.
-
-Apply this diff:
-```diff
-{EDIT_DIFF}```
-
-Final expected file content:
-```rust
-{EXPECTED_CONTENT}```
-"""
-
-VIM_BENCHMARK = BenchmarkSpec(
-    description="Benchmark edit tool in vim mode across models with simple edit tasks.",
-    workspace_prefix="vim-benchmark",
-    tools=("edit", "read"),
-    env={"PI_EDIT_VARIANT": "vim", "PI_STRICT_EDIT_MODE": "1"},
-    initial_prompt=EDIT_PROMPT,
-    retry_instruction="Please try again using the edit tool in vim mode.",
-)
-
-
-def main() -> int:
-    return run_benchmark_main(VIM_BENCHMARK)
-
-
-if __name__ == "__main__":
-    raise SystemExit(main())