rlmgrep 0.1.0__tar.gz → 0.1.1__tar.gz

{rlmgrep-0.1.0 → rlmgrep-0.1.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: rlmgrep
-Version: 0.1.0
+Version: 0.1.1
 Summary: Grep-shaped CLI search powered by DSPy RLM
 Author: rlmgrep
 License: MIT
@@ -17,7 +17,7 @@ Grep-shaped search powered by DSPy RLM. It accepts a natural-language query, sca
 ## Quickstart
 
 ```sh
-uv tool install --python 3.11 .
+uv tool install --python 3.11 rlmgrep
 # or from GitHub:
 # uv tool install --python 3.11 git+https://github.com/halfprice06/rlmgrep.git
 
@@ -71,6 +71,7 @@ Common options:
 - `--type T` include file types (repeatable, comma-separated)
 - `--no-recursive` do not recurse directories
 - `-a`, `--text` treat binary files as text
+- `-y`, `--yes` skip file count confirmation
 - `--model`, `--sub-model` override model names
 - `--api-key`, `--api-base`, `--model-type` override provider settings
 - `--max-iterations`, `--max-llm-calls` cap RLM search effort
@@ -99,6 +100,7 @@ cat README.md | rlmgrep "install"
 - `-g/--glob` matches path globs against normalized paths (forward slashes).
 - Paths are printed relative to the current working directory when possible.
 - If no paths are provided, rlmgrep reads from stdin and uses the synthetic path `<stdin>`; if stdin is empty, it exits with code 2.
+- rlmgrep asks for confirmation when more than 200 files would be loaded (use `-y/--yes` to skip), and aborts when more than 1000 files would be loaded.
 
 ## Output contract (stable for agents)
 
@@ -133,6 +135,8 @@ temperature = 1.0
 max_tokens = 64000
 max_iterations = 10
 max_llm_calls = 20
+file_warn_threshold = 200
+file_hard_max = 1000
 # markitdown_enable_images = false
 # markitdown_image_llm_model = "gpt-5-mini"
 # markitdown_image_llm_provider = "openai"
@@ -168,10 +172,8 @@ If more than one provider key is set and the model does not make the provider ob
 
 - Prefer narrow corpora (globs/types) to reduce token usage.
 - Use `--max-llm-calls` to cap costs; combine with small `--max-iterations` for safety.
-- Always read stderr for warnings (skipped files, config issues, ambiguous API keys).
 - For reproducible parsing, use `-n -H` and avoid context (`-C/-A/-B`).
-- RLM results are verified against real file lines; invalid or duplicate matches are dropped and reported.
-
+  
 ## Development
 
 - Install locally: `pip install -e .` or `uv tool install .`

{rlmgrep-0.1.0 → rlmgrep-0.1.1}/README.md

@@ -5,7 +5,7 @@ Grep-shaped search powered by DSPy RLM. It accepts a natural-language query, sca
 ## Quickstart
 
 ```sh
-uv tool install --python 3.11 .
+uv tool install --python 3.11 rlmgrep
 # or from GitHub:
 # uv tool install --python 3.11 git+https://github.com/halfprice06/rlmgrep.git
 
@@ -59,6 +59,7 @@ Common options:
 - `--type T` include file types (repeatable, comma-separated)
 - `--no-recursive` do not recurse directories
 - `-a`, `--text` treat binary files as text
+- `-y`, `--yes` skip file count confirmation
 - `--model`, `--sub-model` override model names
 - `--api-key`, `--api-base`, `--model-type` override provider settings
 - `--max-iterations`, `--max-llm-calls` cap RLM search effort
@@ -87,6 +88,7 @@ cat README.md | rlmgrep "install"
 - `-g/--glob` matches path globs against normalized paths (forward slashes).
 - Paths are printed relative to the current working directory when possible.
 - If no paths are provided, rlmgrep reads from stdin and uses the synthetic path `<stdin>`; if stdin is empty, it exits with code 2.
+- rlmgrep asks for confirmation when more than 200 files would be loaded (use `-y/--yes` to skip), and aborts when more than 1000 files would be loaded.
 
 ## Output contract (stable for agents)
 
@@ -121,6 +123,8 @@ temperature = 1.0
 max_tokens = 64000
 max_iterations = 10
 max_llm_calls = 20
+file_warn_threshold = 200
+file_hard_max = 1000
 # markitdown_enable_images = false
 # markitdown_image_llm_model = "gpt-5-mini"
 # markitdown_image_llm_provider = "openai"
@@ -156,10 +160,8 @@ If more than one provider key is set and the model does not make the provider ob
 
 - Prefer narrow corpora (globs/types) to reduce token usage.
 - Use `--max-llm-calls` to cap costs; combine with small `--max-iterations` for safety.
-- Always read stderr for warnings (skipped files, config issues, ambiguous API keys).
 - For reproducible parsing, use `-n -H` and avoid context (`-C/-A/-B`).
-- RLM results are verified against real file lines; invalid or duplicate matches are dropped and reported.
-
+  
 ## Development
 
 - Install locally: `pip install -e .` or `uv tool install .`

{rlmgrep-0.1.0 → rlmgrep-0.1.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "rlmgrep"
-version = "0.1.0"
+version = "0.1.1"
 description = "Grep-shaped CLI search powered by DSPy RLM"
 readme = "README.md"
 requires-python = ">=3.11"

{rlmgrep-0.1.0 → rlmgrep-0.1.1}/rlmgrep/cli.py

@@ -8,7 +8,7 @@ from pathlib import Path
 import dspy
 from .config import ensure_default_config, load_config
 from .file_map import build_file_map
-from .ingest import FileRecord, load_files, resolve_type_exts
+from .ingest import FileRecord, collect_candidates, load_files, resolve_type_exts
 from .rlm import Match, build_lm, run_rlm
 from .render import render_matches
 
@@ -17,6 +17,23 @@ def _warn(msg: str) -> None:
     print(f"rlmgrep: {msg}", file=sys.stderr)
 
 
+def _confirm_over_limit(count: int, threshold: int) -> bool:
+    prompt = (
+        f"rlmgrep: {count} files to load (over {threshold}). Continue? [y/N] "
+    )
+    try:
+        with open("/dev/tty", "r+") as tty:
+            print(prompt, file=tty, end="", flush=True)
+            response = tty.readline()
+    except Exception:
+        if not sys.stdin.isatty():
+            _warn("refusing to prompt for confirmation; use --yes to proceed")
+            return False
+        print(prompt, file=sys.stderr, end="", flush=True)
+        response = sys.stdin.readline()
+    return response.strip().lower() in {"y", "yes"}
+
+
 def verify_matches(
     matches: list[Match],
     files: dict[str, FileRecord],
@@ -65,6 +82,7 @@ def _parse_args(argv: list[str]) -> argparse.Namespace:
     parser.add_argument("-m", dest="max_count", type=int, default=None, help="Max matching lines per file")
     parser.add_argument("-a", "--text", dest="binary_as_text", action="store_true", help="Search binary files as text")
     parser.add_argument("--answer", action="store_true", help="Print a narrative answer before grep output")
+    parser.add_argument("-y", "--yes", action="store_true", help="Skip file count confirmation")
 
     parser.add_argument("-g", "--glob", dest="globs", action="append", default=[], help="Include files matching glob (may repeat)")
     parser.add_argument("--type", dest="types", action="append", default=[], help="Include file types (py, js, md, etc.). May repeat")
@@ -328,12 +346,39 @@ def main(argv: list[str] | None = None) -> int:
         }
         warnings: list[str] = []
     else:
-        files, warnings = load_files(
+        warn_threshold = _parse_num(
+            _pick(None, config, "file_warn_threshold", 200), int
+        )
+        hard_max = _parse_num(_pick(None, config, "file_hard_max", 1000), int)
+        if warn_threshold is not None and warn_threshold <= 0:
+            warn_threshold = None
+        if hard_max is not None and hard_max <= 0:
+            hard_max = None
+
+        candidates = collect_candidates(
             args.paths,
             cwd=cwd,
             recursive=args.recursive,
             include_globs=globs,
             type_exts=type_exts,
+        )
+        candidate_count = len(candidates)
+        if hard_max is not None and candidate_count > hard_max:
+            _warn(
+                f"{candidate_count} files to load (over {hard_max}); aborting"
+            )
+            return 2
+        if (
+            warn_threshold is not None
+            and candidate_count > warn_threshold
+            and not args.yes
+        ):
+            if not _confirm_over_limit(candidate_count, warn_threshold):
+                return 2
+
+        files, warnings = load_files(
+            candidates,
+            cwd=cwd,
             markitdown=markitdown,
             enable_images=md_enable_images,
             enable_audio=md_enable_audio,

{rlmgrep-0.1.0 → rlmgrep-0.1.1}/rlmgrep/config.py

@@ -19,6 +19,8 @@ DEFAULT_CONFIG_TEXT = "\n".join(
         "max_tokens = 64000",
         "max_iterations = 10",
         "max_llm_calls = 20",
+        "file_warn_threshold = 200",
+        "file_hard_max = 1000",
         "# markitdown_enable_images = false",
         "# markitdown_image_llm_model = \"gpt-5-mini\"",
         "# markitdown_image_llm_provider = \"openai\"",

{rlmgrep-0.1.0 → rlmgrep-0.1.1}/rlmgrep/ingest.py

@@ -237,12 +237,34 @@ def _matches_globs(path: str, globs: list[str]) -> bool:
     return False
 
 
-def load_files(
+def collect_candidates(
     paths: Iterable[str],
     cwd: Path,
     recursive: bool = True,
     include_globs: list[str] | None = None,
     type_exts: set[str] | None = None,
+) -> list[Path]:
+    files = collect_files(paths, recursive=recursive)
+    candidates: list[Path] = []
+    for fp in files:
+        try:
+            key = fp.relative_to(cwd).as_posix()
+        except ValueError:
+            key = fp.as_posix()
+
+        if include_globs and not _matches_globs(key, include_globs):
+            continue
+
+        if type_exts and fp.suffix.lower() not in type_exts:
+            continue
+
+        candidates.append(fp)
+    return candidates
+
+
+def load_files(
+    candidates: Iterable[Path],
+    cwd: Path,
     markitdown: Any | None = None,
     enable_images: bool = False,
     enable_audio: bool = False,
@@ -254,20 +276,12 @@ def load_files(
     image_convert_count = 0
     audio_convert_count = 0
 
-    files = collect_files(paths, recursive=recursive)
-    for fp in files:
+    for fp in candidates:
         try:
             key = fp.relative_to(cwd).as_posix()
         except ValueError:
             key = fp.as_posix()
 
-        if include_globs and not _matches_globs(key, include_globs):
-            continue
-
-        if type_exts:
-            if fp.suffix.lower() not in type_exts:
-                continue
-
         suffix = fp.suffix.lower()
         if markitdown is not None and not binary_as_text:
             if enable_images and suffix in IMAGE_EXTS:

{rlmgrep-0.1.0 → rlmgrep-0.1.1}/rlmgrep.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: rlmgrep
-Version: 0.1.0
+Version: 0.1.1
 Summary: Grep-shaped CLI search powered by DSPy RLM
 Author: rlmgrep
 License: MIT
@@ -17,7 +17,7 @@ Grep-shaped search powered by DSPy RLM. It accepts a natural-language query, sca
 ## Quickstart
 
 ```sh
-uv tool install --python 3.11 .
+uv tool install --python 3.11 rlmgrep
 # or from GitHub:
 # uv tool install --python 3.11 git+https://github.com/halfprice06/rlmgrep.git
 
@@ -71,6 +71,7 @@ Common options:
 - `--type T` include file types (repeatable, comma-separated)
 - `--no-recursive` do not recurse directories
 - `-a`, `--text` treat binary files as text
+- `-y`, `--yes` skip file count confirmation
 - `--model`, `--sub-model` override model names
 - `--api-key`, `--api-base`, `--model-type` override provider settings
 - `--max-iterations`, `--max-llm-calls` cap RLM search effort
@@ -99,6 +100,7 @@ cat README.md | rlmgrep "install"
 - `-g/--glob` matches path globs against normalized paths (forward slashes).
 - Paths are printed relative to the current working directory when possible.
 - If no paths are provided, rlmgrep reads from stdin and uses the synthetic path `<stdin>`; if stdin is empty, it exits with code 2.
+- rlmgrep asks for confirmation when more than 200 files would be loaded (use `-y/--yes` to skip), and aborts when more than 1000 files would be loaded.
 
 ## Output contract (stable for agents)
 
@@ -133,6 +135,8 @@ temperature = 1.0
 max_tokens = 64000
 max_iterations = 10
 max_llm_calls = 20
+file_warn_threshold = 200
+file_hard_max = 1000
 # markitdown_enable_images = false
 # markitdown_image_llm_model = "gpt-5-mini"
 # markitdown_image_llm_provider = "openai"
@@ -168,10 +172,8 @@ If more than one provider key is set and the model does not make the provider ob
 
 - Prefer narrow corpora (globs/types) to reduce token usage.
 - Use `--max-llm-calls` to cap costs; combine with small `--max-iterations` for safety.
-- Always read stderr for warnings (skipped files, config issues, ambiguous API keys).
 - For reproducible parsing, use `-n -H` and avoid context (`-C/-A/-B`).
-- RLM results are verified against real file lines; invalid or duplicate matches are dropped and reported.
-
+  
 ## Development
 
 - Install locally: `pip install -e .` or `uv tool install .`