diff --git a/CHANGELOG.md b/CHANGELOG.md
index c9aab9b..ff50979 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,13 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+- `conflict_markers` check: fails on leftover git merge conflict markers
+  (`<<<<<<<`, `>>>>>>>`, `|||||||`).
+- `becwright init --from-claude-md` expands a broad "good practices" / "buenas
+  prácticas" phrase into the deterministic code-hygiene rule set (no secrets,
+  `eval`, debug leftovers, or conflict markers), language-aware and de-duplicated.
+
 ## [0.3.0] — 2026-07-01
 
 ### Added
diff --git a/README.es.md b/README.es.md
index 84d53b3..22786c7 100644
--- a/README.es.md
+++ b/README.es.md
@@ -306,6 +306,7 @@ atar cada regla a su *por qué*.
 | `hardcoded_secrets` | Claves AWS, claves privadas, `password = "..."` literales | cualquiera | `blocking` |
 | `debug_remnants` | `breakpoint()`, `pdb.set_trace()`, `import pdb` olvidados | Python | `blocking` |
 | `dangerous_eval` | Llamadas a `eval()` / `exec()` | cualquiera | `blocking` |
+| `conflict_markers` | Marcadores de conflicto de merge olvidados (`<<<<<<<`) | cualquiera | `blocking` |
 | `wildcard_imports` | `from x import *` | Python | `warning` |
 
 ## Reglas listas para usar (sin escribir nada)
diff --git a/README.md b/README.md
index 570127d..4db2614 100644
--- a/README.md
+++ b/README.md
@@ -296,6 +296,7 @@ may miss exotic cases, and the real value is in tying each rule to its *why*.
 | `hardcoded_secrets` | AWS keys, private keys, `password = "..."` literals | any | `blocking` |
 | `debug_remnants` | Forgotten `breakpoint()`, `pdb.set_trace()`, `import pdb` | Python | `blocking` |
 | `dangerous_eval` | `eval()` / `exec()` calls | any | `blocking` |
+| `conflict_markers` | Leftover git merge conflict markers (`<<<<<<<`) | any | `blocking` |
 | `wildcard_imports` | `from x import *` | Python | `warning` |
 
 Example rules to copy into your `.bec/rules.yaml`:
diff --git a/documentation/usage.es.md b/documentation/usage.es.md
index 52a897d..567f0bb 100644
--- a/documentation/usage.es.md
+++ b/documentation/usage.es.md
@@ -41,7 +41,9 @@ a mano: `becwright install` más un `.bec/rules.yaml` que escribas vos.)
 > `debugger`, `console.log`, breakpoints, imports con `*`, tokens en logs — y
 > convierte cada una en una regla enforzable, informando qué frase la disparó.
 > También detecta un límite de líneas por archivo ("archivos < 800 líneas" →
-> `max_lines`), ignorando reglas de largo de función que no puede enforzar. Es
+> `max_lines`), ignorando reglas de largo de función que no puede enforzar. Una
+> frase amplia como "seguir buenas prácticas" se expande al set determinista de
+> higiene (sin secretos, `eval`, restos de debug ni marcadores de conflicto). Es
 > best-effort y según el lenguaje, así que **revisá el resultado**; lo de criterio
 > (arquitectura, naming, inmutabilidad) no tiene check determinista y se queda en
 > `CLAUDE.md`. Combinalo con `--baseline` para adoptar en un repo sucio de una.
diff --git a/documentation/usage.md b/documentation/usage.md
index c82f94f..bcc57ce 100644
--- a/documentation/usage.md
+++ b/documentation/usage.md
@@ -39,7 +39,9 @@ From then on, every `git commit` runs the checks. (You can also set up by hand:
 > `console.log`, breakpoints, wildcard imports, tokens in logs — and turns each
 > into an enforceable rule, reporting which phrase matched. It also picks up a
 > per-file line cap ("files under 800 lines" → `max_lines`), ignoring
-> function-length rules it can't enforce. This is best-effort and language-aware,
+> function-length rules it can't enforce. A broad phrase like "follow good
+> practices" expands to the deterministic hygiene set (no secrets, `eval`, debug
+> leftovers, or merge-conflict markers). This is best-effort and language-aware,
 > so **review the result**; judgment-based guidance (architecture, naming,
 > immutability) has no deterministic check and stays in `CLAUDE.md`. Combine with
 > `--baseline` to adopt on a dirty repo in one step.
diff --git a/src/becwright/checks/conflict_markers.py b/src/becwright/checks/conflict_markers.py
new file mode 100644
index 0000000..ba71f07
--- /dev/null
+++ b/src/becwright/checks/conflict_markers.py
@@ -0,0 +1,39 @@
+from __future__ import annotations
+
+import re
+import sys
+
+from ._ignore import is_ignored
+
+# The angle/pipe markers a merge conflict leaves behind. `=======` is omitted on
+# purpose: it also appears in legitimate text (e.g. a Markdown heading underline),
+# and every real conflict already carries `<<<<<<<` and `>>>>>>>`, so it is caught.
+PATTERN = re.compile(r"^(?:<{7}|>{7}|\|{7})(?:\s|$)")
+
+
+def find_violations(paths: list[str]) -> list[tuple[str, int, str]]:
+    violations: list[tuple[str, int, str]] = []
+    for path in paths:
+        path = path.strip()
+        if not path:
+            continue
+        try:
+            with open(path, encoding="utf-8") as f:
+                for lineno, line in enumerate(f, start=1):
+                    if PATTERN.search(line) and not is_ignored(line):
+                        violations.append((path, lineno, line.rstrip()))
+        except (FileNotFoundError, IsADirectoryError, UnicodeDecodeError):
+            continue
+    return violations
+
+
+def main() -> int:
+    violations = find_violations(sys.stdin.read().splitlines())
+    for path, lineno, line in violations:
+        print(f"  {path}:{lineno}")
+        print(f"      > {line}")
+    return 1 if violations else 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/src/becwright/cli.py b/src/becwright/cli.py
index 4b60ef7..226dcb7 100644
--- a/src/becwright/cli.py
+++ b/src/becwright/cli.py
@@ -126,6 +126,7 @@ def _cmd_uninstall(_: argparse.Namespace) -> int:
     "max_lines": "fail if a file exceeds --max lines (any language)",
     "require": "fail if a regex (--pattern) is missing from the files (any language)",
     "filename": "fail on file names matching --forbid or not matching --require (any language)",
+    "conflict_markers": "leftover git merge conflict markers (any language)",
 }
 
 
@@ -336,6 +337,22 @@ def _max_lines_cap(text: str) -> int | None:
     return cap if 50 <= cap <= 5000 else None
 
 
+# "Good practices" / "clean commits" is too broad to map to one check, so it
+# expands to the deterministic hygiene subset of the signals above (plus conflict
+# markers). Opinionated or narrow signals (wildcard imports, tokens in logs) are
+# left out — they only fire on their own explicit mention.
+_HYGIENE_IDS = frozenset({
+    "no-hardcoded-secrets", "no-dangerous-eval", "no-debug-remnants",
+    "no-debugger-js", "no-console-log-js",
+})
+_GOOD_PRACTICES_TRIGGERS = (
+    "good practices", "best practices", "buenas practicas", "buenas prácticas",
+    "mejores practicas", "mejores prácticas", "clean commit", "commit hygiene",
+)
+_CONFLICT_TRIGGERS = ("conflict marker", "merge conflict",
+                      "marcador de conflicto", "conflicto de merge")
+
+
 def _rules_from_claude_md(text: str, langs: list[str]) -> list[tuple[dict, str]]:
     """Best-effort mapping from prohibitions written in CLAUDE.md to executable
     rules becwright can actually enforce. Returns (rule_dict, matched_trigger)
@@ -347,6 +364,7 @@ def _rules_from_claude_md(text: str, langs: list[str]) -> list[tuple[dict, str]]
         "python": ["**/*.py"] if "python" in langs else [],
         "jsts": [g for g in source if g.endswith((".js", ".ts"))],
     }
+    good = any(t in lowered for t in _GOOD_PRACTICES_TRIGGERS)
     derived: list[tuple[dict, str]] = []
     for signal in _CLAUDE_SIGNALS:
         paths = paths_for[signal["lang"]]
@@ -354,12 +372,24 @@ def _rules_from_claude_md(text: str, langs: list[str]) -> list[tuple[dict, str]]
             continue
         trigger = next((t for t in signal["triggers"] if t in lowered), None)
         if trigger is None:
-            continue
+            if good and signal["id"] in _HYGIENE_IDS:
+                trigger = "good practices"
+            else:
+                continue
         derived.append((dict(
             id=signal["id"], paths=paths, check=signal["check"],
             intent=signal["intent"], why=signal["why"], severity=signal["severity"],
         ), trigger))
 
+    conflict_trigger = next((t for t in _CONFLICT_TRIGGERS if t in lowered), None)
+    if good or conflict_trigger:
+        derived.append((dict(
+            id="no-conflict-markers", paths=["**/*"], severity="blocking",
+            check="becwright run conflict_markers",
+            intent="Merge conflict markers must never be committed.",
+            why="A committed conflict marker (<<<<<<<) means unmerged code shipped."),
+            conflict_trigger or "good practices"))
+
     cap = _max_lines_cap(text)
     if cap is not None and source:
         derived.append((dict(
diff --git a/tests/test_init.py b/tests/test_init.py
index 5ad6194..3bba224 100644
--- a/tests/test_init.py
+++ b/tests/test_init.py
@@ -322,6 +322,29 @@ def test_rules_from_claude_md_maps_file_line_cap():
     assert by_id["max-file-lines"]["severity"] == "warning"
 
 
+def test_rules_from_claude_md_good_practices_expands_hygiene():
+    derived = cli._rules_from_claude_md("Always follow good practices.", ["python", "ts"])
+    ids = {r["id"] for r, _ in derived}
+    assert {"no-hardcoded-secrets", "no-dangerous-eval", "no-debug-remnants",
+            "no-debugger-js", "no-console-log-js", "no-conflict-markers"} <= ids
+    # opinionated/narrow signals are not pulled in by the broad phrase
+    assert "no-wildcard-imports" not in ids and "no-token-in-logs" not in ids
+
+
+def test_rules_from_claude_md_conflict_markers_on_explicit_mention():
+    ids = {r["id"] for r, _ in cli._rules_from_claude_md(
+        "Never commit merge conflict markers.", ["python"])}
+    assert ids == {"no-conflict-markers"}
+
+
+def test_rules_from_claude_md_good_practices_keeps_specific_trigger():
+    # a specific mention wins the trigger label, and the rule is not duplicated
+    derived = cli._rules_from_claude_md(
+        "No hardcoded secrets. Follow good practices.", ["python"])
+    secrets = [(r, t) for r, t in derived if r["id"] == "no-hardcoded-secrets"]
+    assert len(secrets) == 1 and secrets[0][1] == "hardcod"
+
+
 def test_init_from_claude_md_composes_with_baseline(tmp_path, monkeypatch):
     monkeypatch.setenv("PYTHONPATH", str(_SRC))
     _init_repo(tmp_path)
diff --git a/tests/test_metric_naming_checks.py b/tests/test_metric_naming_checks.py
index f182b7c..37ddaeb 100644
--- a/tests/test_metric_naming_checks.py
+++ b/tests/test_metric_naming_checks.py
@@ -1,6 +1,6 @@
 import io
 
-from becwright.checks import filename, max_lines, require
+from becwright.checks import conflict_markers, filename, max_lines, require
 
 
 def _write(tmp_path, name, text):
@@ -98,3 +98,24 @@ def test_filename_main_flags_and_exits_one(tmp_path, monkeypatch, capsys):
     monkeypatch.setattr("sys.stdin", io.StringIO(bad + "\n"))
     assert filename.main(["--forbid", "tipos"]) == 1
     assert "forbidden pattern" in capsys.readouterr().out
+
+
+# --- conflict_markers ---
+
+def test_conflict_markers_flags_angle_markers(tmp_path):
+    f = _write(tmp_path, "a.py", "x = 1\n<<<<<<< HEAD\ny = 2\n>>>>>>> branch\n")
+    v = conflict_markers.find_violations([f])
+    assert {ln for _p, ln, _l in v} == {2, 4}
+
+
+def test_conflict_markers_ignores_markdown_underline(tmp_path):
+    # `=======` is a legit Markdown h1 underline; only angle/pipe markers are flagged.
+    f = _write(tmp_path, "r.md", "Title\n=======\nbody\n")
+    assert conflict_markers.find_violations([f]) == []
+
+
+def test_conflict_markers_main(tmp_path, monkeypatch, capsys):
+    f = _write(tmp_path, "a.py", "||||||| base\n")
+    monkeypatch.setattr("sys.stdin", io.StringIO(f + "\n"))
+    assert conflict_markers.main() == 1
+    assert f in capsys.readouterr().out