fix(config): one bounded backups/config/ dir replaces four config.yaml.bak schemes

Four writers each dropped their own uniquely-named copy of config.yaml next to
the real file and none of them ever deleted anything: hermes setup
(config.yaml.bak.YYYYMMDD_HHMMSS, one per run even with no change), the
corrupt-YAML snapshot (config.yaml.corrupt.<ts>.bak), hermes migrate xai
(config.yaml.bak-pre-migrate-xai-<ts>) and the Docker boot migration
(config.yaml.bak-<ts>, .env.bak-<ts>). A home dir accumulated a dozen variants
with no way to tell which mattered.

hermes_cli/config_backups.py::backup_config is now the single writer:
backups/config/config.yaml.<reason>.<YYYYMMDD-HHMMSS>, skipped when the newest
copy for that reason is byte-identical, rotated to the newest five per reason.
backups/ is already excluded from full backups so nothing nests. Legacy
siblings written by the old schemes are moved into the dir on first use;
hand-named copies (config.yaml.bak-my-note) are left alone.

Live: three `hermes setup --non-interactive` runs against an unchanged config
went from three .bak files in HERMES_HOME to one pre-setup copy under
backups/config/; repeated loads of broken YAML produce one corrupt copy
instead of one per process (deduped by content).

hermes_cli/config.py

@@ -49,35 +49,6 @@ class InvalidUserConfigError(RuntimeError):
     """Raised when a run that cannot repair config finds invalid user YAML."""
 
 
-def _backup_corrupt_config(config_path: Path) -> Optional[Path]:
-    """Copy an unparseable ``config.yaml`` to a timestamped ``.corrupt.*.bak``; None on skip/failure.
-    Symlinks are not followed (never clobber whatever a malicious symlink points at). A sibling
-    backup of the same size means this corruption was already snapshotted — skip to avoid churn.
-
-    Returns the backup path on success, else ``None``. See #21541.
-    """
-    try:
-        if config_path.is_symlink():
-            return None
-        st = config_path.stat()
-        if st.st_size == 0:
-            return None
-        ts = time.strftime("%Y%m%d-%H%M%S")
-        backup_path = config_path.with_name(f"{config_path.name}.corrupt.{ts}.bak")
-        for existing in config_path.parent.glob(f"{config_path.name}.corrupt.*.bak"):
-            try:
-                if existing.stat().st_size == st.st_size:
-                    return None
-            except OSError:
-                continue
-        if backup_path.exists():
-            return None
-        shutil.copy2(config_path, backup_path)
-        return backup_path
-    except Exception:
-        return None
-
-
 _PARSE_FAILURE_FALLBACK_MSG = {
     "last-known-good": (
         "Keeping the previously loaded config for this process — "
@@ -108,7 +79,8 @@ def _warn_config_parse_failure(
     if key in _CONFIG_PARSE_WARNED:
         return
     _CONFIG_PARSE_WARNED.add(key)
-    backup_path = _backup_corrupt_config(config_path)
+    from hermes_cli.config_backups import backup_config
+    backup_path = backup_config(config_path, "corrupt")
     msg = f"Failed to parse {config_path}: {exc}. " + _PARSE_FAILURE_FALLBACK_MSG.get(
         fallback, _PARSE_FAILURE_DEFAULTS_MSG)
     if backup_path is not None:
@@ -515,7 +487,8 @@ def require_parseable_user_config(*, ignore_user_config: bool = False) -> None:
             return
         parse_error = TypeError(f"top-level YAML value must be a mapping, got {type(data).__name__}")
 
-    backup_path = _backup_corrupt_config(config_path)
+    from hermes_cli.config_backups import backup_config
+    backup_path = backup_config(config_path, "corrupt")
     message = (
         f"Refusing non-interactive startup because {config_path} is invalid: "
         f"{parse_error}. Repair the file or pass --ignore-user-config to "
@@ -1944,7 +1917,7 @@ def _refuse_overwrite(config_path: Path, reason: str, exc: Exception, fix: str)
 
 
 _FIX_PERMS = "Fix the file permissions or move it aside first."
-_FIX_YAML = "Fix the file or restore from a .corrupt.*.bak backup first."
+_FIX_YAML = "Fix the file or restore a copy from backups/config/ first."
 
 
 def require_readable_config_before_write(config_path: Optional[Path] = None) -> Dict[str, Any]:
@@ -1977,7 +1950,7 @@ def require_readable_config_before_write(config_path: Optional[Path] = None) ->
         _warn_config_parse_failure(config_path, exc, fallback="refuse-write")
         raise RuntimeError(
             f"Refusing to overwrite {config_path}: top-level YAML must be a mapping, got "
-            f"{type(loaded).__name__}. Fix the file or restore from a .corrupt.*.bak backup first."
+            f"{type(loaded).__name__}. Fix the file or restore a copy from backups/config/ first."
         ) from exc
     return loaded
 

hermes_cli/config_backups.py

@@ -0,0 +1,80 @@
+"""Point-in-time copies of ``config.yaml``: one directory, one naming scheme, bounded count.
+
+Every writer that wants a "before" copy of the user's config (setup wizard, corrupt-file
+snapshot, model migrations) goes through :func:`backup_config`. Copies live in
+``<HERMES_HOME>/backups/config/`` — ``backups/`` is already excluded from full backups, so they
+never nest — as ``config.yaml.<reason>.<YYYYMMDD-HHMMSS>``. A copy identical to the newest one
+for the same reason is skipped, and only the newest ``keep`` per reason survive, so repeated
+``hermes setup`` runs or a gateway restarting against broken YAML cannot litter the home dir.
+"""
+
+from __future__ import annotations
+
+import filecmp
+import logging
+import shutil
+import time
+from pathlib import Path
+from typing import Optional
+
+logger = logging.getLogger(__name__)
+
+BACKUPS_SUBDIR = Path("backups") / "config"
+DEFAULT_KEEP = 5
+
+# Names earlier code wrote next to config.yaml (setup wizard, corrupt snapshot, xai migration).
+# Moved into the backups dir on first use so they stop accumulating in the home root; hand-named
+# copies (``config.yaml.bak-my-note``) are the user's and are never touched.
+_LEGACY_SIBLING_GLOBS = ("config.yaml.bak.[0-9]*", "config.yaml.corrupt.*.bak", "config.yaml.bak-pre-migrate-*")
+
+
+def backups_dir(config_path: Path) -> Path:
+    return config_path.parent / BACKUPS_SUBDIR
+
+
+def list_config_backups(config_path: Path, reason: Optional[str] = None) -> list[Path]:
+    """Existing backups, newest first; filtered to one *reason* when given."""
+    root = backups_dir(config_path)
+    if not root.is_dir():
+        return []
+    prefix = f"{config_path.name}.{reason}." if reason else f"{config_path.name}."
+    return sorted((p for p in root.iterdir() if p.is_file() and p.name.startswith(prefix)),
+                  key=lambda p: p.name, reverse=True)
+
+
+def backup_config(config_path: Path, reason: str, *, keep: int = DEFAULT_KEEP) -> Optional[Path]:
+    """Copy *config_path* to the backups dir; return the new path, or None when skipped/failed.
+
+    Skips when the file is missing/empty, or when the newest backup for *reason* already holds
+    identical bytes. Never raises: a failed backup must not block the write it precedes.
+    """
+    try:
+        if not config_path.is_file() or config_path.stat().st_size == 0:
+            return None
+        root = backups_dir(config_path)
+        root.mkdir(parents=True, exist_ok=True)
+        _sweep_legacy_siblings(config_path, root)
+        existing = list_config_backups(config_path, reason)
+        if existing and filecmp.cmp(config_path, existing[0], shallow=False):
+            return None
+        dest = root / f"{config_path.name}.{reason}.{time.strftime('%Y%m%d-%H%M%S')}"
+        if dest.is_symlink() or dest.exists():  # never write through a planted link
+            return None
+        shutil.copy2(config_path, dest)
+        for stale in [dest, *existing][keep:]:
+            stale.unlink(missing_ok=True)
+        return dest
+    except OSError as exc:
+        logger.warning("Could not back up %s (%s): %s", config_path, reason, exc)
+        return None
+
+
+def _sweep_legacy_siblings(config_path: Path, root: Path) -> None:
+    for pattern in _LEGACY_SIBLING_GLOBS:
+        for old in config_path.parent.glob(pattern):
+            if not old.is_file() or old.is_symlink():
+                continue
+            try:
+                old.replace(root / old.name)
+            except OSError as exc:
+                logger.debug("Could not move legacy backup %s: %s", old, exc)

hermes_cli/setup.py

@@ -580,20 +580,6 @@ def run_setup_wizard(args):
             return None
 
 
-def _backup_config_file(config_path: Path) -> Path | None:
-    """Back up config.yaml before setup modifies it; None when absent or copy fails."""
-    if not config_path.exists():
-        return None
-    import shutil
-    from datetime import datetime
-    backup_path = config_path.with_suffix(f".yaml.bak.{datetime.now().strftime('%Y%m%d_%H%M%S')}")
-    try:
-        shutil.copy2(config_path, backup_path)
-        return backup_path
-    except Exception:
-        return None
-
-
 def _run_setup_section(config: dict, section: str) -> None:
     """``hermes setup <section>``: run one SETUP_SECTIONS entry under the banner."""
     entry = next(((label, func) for key, label, func in SETUP_SECTIONS if key == section), None)
@@ -672,7 +658,8 @@ def _run_setup_wizard_impl(args):
     hermes_home = get_hermes_home()
     # Back up existing config before setup modifies it (#3522)
     config_path = get_config_path()
-    _backup_path = _backup_config_file(config_path)
+    from hermes_cli.config_backups import backup_config
+    _backup_path = backup_config(config_path, "pre-setup")
 
     # Non-interactive environments (headless SSH, Docker, CI/CD)
     if getattr(args, 'non_interactive', False) or not is_interactive_stdin():

hermes_cli/xai_retirement.py

@@ -1,9 +1,7 @@
 """Detect xAI models retired on May 15, 2026 and migrate config.yaml references."""
 from __future__ import annotations
 
-import datetime as _dt
 import io
-import shutil
 from dataclasses import dataclass
 from pathlib import Path
 from typing import Any, Dict, List, Optional
@@ -130,7 +128,7 @@ def apply_migration(
     config_path: Path, issues: List[RetirementIssue], backup: bool = True) -> ApplyResult:
     """Rewrite ``config_path`` in place (ruamel round-trip: comments, order, type literals kept).
 
-    Unless ``backup=False`` a copy goes to ``<config_path>.bak-pre-migrate-xai-YYYYMMDD-HHMMSS``.
+    Unless ``backup=False`` a copy goes to ``backups/config/`` (reason ``pre-migrate-xai``).
     """
     from ruamel.yaml import YAML  # local import — avoid hard dep at module load
     config_path = Path(config_path)
@@ -162,9 +160,8 @@ def apply_migration(
 
     backup_path: Optional[Path] = None
     if backup:
-        ts = _dt.datetime.now().strftime("%Y%m%d-%H%M%S")
+        from hermes_cli.config_backups import backup_config
-        backup_path = config_path.with_name(f"{config_path.name}.bak-pre-migrate-xai-{ts}")
+        backup_path = backup_config(config_path, "pre-migrate-xai")
-        shutil.copy2(config_path, backup_path)
 
     from hermes_cli.config import require_readable_config_before_write
     from utils import atomic_write_text

scripts/docker_config_migrate.py

@@ -4,7 +4,6 @@ from __future__ import annotations
 
 import shutil
 import sys
-from datetime import datetime, timezone
 from pathlib import Path
 from typing import Iterable
 
@@ -14,6 +13,7 @@ from hermes_cli.config import (
     get_env_path,
     migrate_config,
 )
+from hermes_cli.config_backups import backup_config, list_config_backups
 from hermes_cli.config_migrations import (
     SUPPORT_FLOOR_VERSION,
     support_floor_message,
@@ -21,26 +21,14 @@ from hermes_cli.config_migrations import (
 from utils import env_var_enabled
 
 
-def _backup_path(path: Path, stamp: str) -> Path:
-    base = path.with_name(f"{path.name}.bak-{stamp}")
-    if not base.exists():
-        return base
-    for index in range(1, 1000):
-        candidate = path.with_name(f"{path.name}.bak-{stamp}.{index}")
-        if not candidate.exists():
-            return candidate
-    raise RuntimeError(f"could not choose a backup path for {path}")
-
-
 def _backup_existing(paths: Iterable[Path]) -> dict[Path, Path]:
-    stamp = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
+    """Snapshot each file into backups/config/; an identical existing snapshot is reused."""
     backups: dict[Path, Path] = {}
     for path in paths:
-        if not path.is_file():
+        dest = backup_config(path, "pre-docker-migrate") or next(
-            continue
+            iter(list_config_backups(path, "pre-docker-migrate")), None)
-        dest = _backup_path(path, stamp)
+        if dest is not None:
-        shutil.copy2(path, dest)
+            backups[path] = dest
-        backups[path] = dest
     return backups
 
 

tests/hermes_cli/test_config.py

@@ -178,7 +178,7 @@ class TestLoadConfigParseFailure:
             load_config()
             err = capsys.readouterr().err
 
-            baks = list(tmp_path.glob("config.yaml.corrupt.*.bak"))
+            baks = list((tmp_path / "backups" / "config").glob("config.yaml.corrupt.*"))
             assert len(baks) == 1, f"expected one backup, got {baks}"
             # Backup preserves the original broken content verbatim
             assert baks[0].read_text() == broken
@@ -331,7 +331,7 @@ class TestSaveAndLoadRoundtrip:
                 set_config_value("model.default", "gpt-4o")
 
         assert config_path.read_text(encoding="utf-8") == original
-        assert list(tmp_path.glob("config.yaml.corrupt.*.bak")), (
+        assert list((tmp_path / "backups" / "config").glob("config.yaml.corrupt.*")), (
             "parse-failure path should snapshot a corrupt backup before refusing"
         )
 
@@ -348,7 +348,7 @@ class TestSaveAndLoadRoundtrip:
 
         assert config_path.read_text(encoding="utf-8") == original
         assert (tmp_path / ".env").read_text(encoding="utf-8") == "TERMINAL_TIMEOUT=30\n"
-        assert list(tmp_path.glob("config.yaml.corrupt.*.bak")), (
+        assert list((tmp_path / "backups" / "config").glob("config.yaml.corrupt.*")), (
             "unset parse-failure path should snapshot a corrupt backup before refusing"
         )
 
@@ -363,7 +363,7 @@ class TestSaveAndLoadRoundtrip:
                 set_config_value("model.default", "gpt-4o")
 
         assert config_path.read_text(encoding="utf-8") == original
-        assert list(tmp_path.glob("config.yaml.corrupt.*.bak")), (
+        assert list((tmp_path / "backups" / "config").glob("config.yaml.corrupt.*")), (
             "non-mapping root should snapshot a corrupt backup before refusing"
         )
 
@@ -378,7 +378,7 @@ class TestSaveAndLoadRoundtrip:
                 unset_config_value("model.default")
 
         assert config_path.read_text(encoding="utf-8") == original
-        assert list(tmp_path.glob("config.yaml.corrupt.*.bak"))
+        assert list((tmp_path / "backups" / "config").glob("config.yaml.corrupt.*"))
 
     def test_config_set_allows_valid_empty_mapping(self, tmp_path):
         """A genuine empty {} config must still be writable (not a false refuse)."""
@@ -403,7 +403,7 @@ class TestSaveAndLoadRoundtrip:
             atomic_config_write(config_path, {"model": {"provider": "openai"}})
 
         assert config_path.read_text(encoding="utf-8") == original
-        assert list(tmp_path.glob("config.yaml.corrupt.*.bak"))
+        assert list((tmp_path / "backups" / "config").glob("config.yaml.corrupt.*"))
 
 class TestSaveEnvValueSecure:
 

tests/hermes_cli/test_config_backups.py

@@ -0,0 +1,40 @@
+"""config.yaml backups: one dir, deduped, bounded — never a pile of siblings in HERMES_HOME."""
+from pathlib import Path
+
+from hermes_cli.config_backups import backup_config, list_config_backups
+
+
+def test_repeat_backups_dedupe_and_rotate(tmp_path: Path, monkeypatch):
+    cfg = tmp_path / "config.yaml"
+    cfg.write_text("model: a\n")
+    stamps = iter(f"2026010100000{i}" for i in range(10))
+    monkeypatch.setattr("hermes_cli.config_backups.time.strftime", lambda _fmt: next(stamps))
+
+    first = backup_config(cfg, "pre-setup", keep=2)
+    assert first is not None and first.parent == tmp_path / "backups" / "config"
+    # Same bytes again → no new file (the hermes-setup-three-times case).
+    assert backup_config(cfg, "pre-setup", keep=2) is None
+    for i in range(3):
+        cfg.write_text(f"model: {i}\n")
+        backup_config(cfg, "pre-setup", keep=2)
+    kept = list_config_backups(cfg, "pre-setup")
+    assert len(kept) == 2 and kept[0].read_text() == "model: 2\n"
+    # Nothing left beside config.yaml in the home root.
+    assert [p.name for p in tmp_path.iterdir() if p.is_file()] == ["config.yaml"]
+
+
+def test_legacy_siblings_move_but_user_named_copies_stay(tmp_path: Path):
+    cfg = tmp_path / "config.yaml"
+    cfg.write_text("model: a\n")
+    for name in ("config.yaml.bak.1778718391", "config.yaml.corrupt.20260729-093706.bak",
+                 "config.yaml.bak-pre-migrate-xai-20260515-120000"):
+        (tmp_path / name).write_text("old")
+    (tmp_path / "config.yaml.bak-my-note").write_text("mine")
+
+    backup_config(cfg, "corrupt")
+
+    root = tmp_path / "backups" / "config"
+    assert (root / "config.yaml.bak.1778718391").exists()
+    assert (root / "config.yaml.corrupt.20260729-093706.bak").exists()
+    assert (tmp_path / "config.yaml.bak-my-note").read_text() == "mine"
+    assert not list(tmp_path.glob("config.yaml.bak.*")) and not list(tmp_path.glob("config.yaml.corrupt.*"))

tests/hermes_cli/test_migrate_xai.py

@@ -162,7 +162,7 @@ class TestBackup:
         result = apply_migration(trap_config, issues, backup=False)
         assert result.backup_path is None
         # No bak file in the directory
-        assert not list(trap_config.parent.glob("*.bak-pre-migrate-xai-*"))
+        assert not list(trap_config.parent.rglob("config.yaml.pre-migrate-xai.*"))
 
 
 

tests/hermes_cli/test_noninteractive_config_guard.py

@@ -61,7 +61,7 @@ def test_noninteractive_guard_rejects_malformed_yaml(args, tmp_path, caplog, cap
         for record in caplog.records
     )
     assert config_path.read_text(encoding="utf-8") == broken
-    backups = list(tmp_path.glob("config.yaml.corrupt.*.bak"))
+    backups = list((tmp_path / "backups" / "config").glob("config.yaml.corrupt.*"))
     assert len(backups) == 1
     assert backups[0].read_text(encoding="utf-8") == broken
 
@@ -131,7 +131,7 @@ def test_explicit_config_bypasses_allow_noninteractive_recovery(args, tmp_path):
     main_mod._guard_noninteractive_user_config(args)
 
     assert args._noninteractive_config_validated is True
-    assert list(tmp_path.glob("config.yaml.corrupt.*.bak")) == []
+    assert list(tmp_path.rglob("config.yaml.corrupt.*")) == []
 
 
 def test_interactive_chat_keeps_existing_repair_behavior(tmp_path):
@@ -143,7 +143,7 @@ def test_interactive_chat_keeps_existing_repair_behavior(tmp_path):
     main_mod._guard_noninteractive_user_config(args)
 
     assert not hasattr(args, "_noninteractive_config_validated")
-    assert list(tmp_path.glob("config.yaml.corrupt.*.bak")) == []
+    assert list(tmp_path.rglob("config.yaml.corrupt.*")) == []
 
 
 @pytest.mark.parametrize(
@@ -163,7 +163,7 @@ def test_queryless_chat_keeps_interactive_repair_behavior(args, tmp_path):
     main_mod._guard_noninteractive_user_config(args)
 
     assert not hasattr(args, "_noninteractive_config_validated")
-    assert list(tmp_path.glob("config.yaml.corrupt.*.bak")) == []
+    assert list(tmp_path.rglob("config.yaml.corrupt.*")) == []
 
 
 def test_env_only_config_bypass_allows_noninteractive_recovery(monkeypatch, tmp_path):
@@ -176,7 +176,7 @@ def test_env_only_config_bypass_allows_noninteractive_recovery(monkeypatch, tmp_
     main_mod._guard_noninteractive_user_config(args)
 
     assert args._noninteractive_config_validated is True
-    assert list(tmp_path.glob("config.yaml.corrupt.*.bak")) == []
+    assert list(tmp_path.rglob("config.yaml.corrupt.*")) == []
 
 
 def test_reused_args_can_retry_after_config_repair(tmp_path):

tests/tools/test_docker_config_migrate.py

@@ -68,8 +68,8 @@ def test_docker_config_migrate_backs_up_and_migrates_legacy_config(tmp_path: Pat
     # max_async_children into max_concurrent_children.
     assert "ttl_hours" not in raw["model_catalog"]
     assert raw["delegation"] == {"max_concurrent_children": 8}
-    assert list(tmp_path.glob("config.yaml.bak-*"))
+    assert list((tmp_path / "backups" / "config").glob("config.yaml.pre-docker-migrate.*"))
-    assert list(tmp_path.glob(".env.bak-*"))
+    assert list((tmp_path / "backups" / "config").glob(".env.pre-docker-migrate.*"))
 
 
 def test_docker_config_migrate_skips_below_floor_config_untouched(tmp_path: Path) -> None:
@@ -170,8 +170,8 @@ def test_docker_config_migrate_restores_backups_after_failed_migration(
 
     assert config_path.read_text(encoding="utf-8") == original_config
     assert env_path.read_text(encoding="utf-8") == original_env
-    assert list(tmp_path.glob("config.yaml.bak-*"))
+    assert list((tmp_path / "backups" / "config").glob("config.yaml.pre-docker-migrate.*"))
-    assert list(tmp_path.glob(".env.bak-*"))
+    assert list((tmp_path / "backups" / "config").glob(".env.pre-docker-migrate.*"))
 
 
 def test_docker_config_migrate_restores_backups_when_version_does_not_advance(
@@ -244,7 +244,7 @@ def test_docker_config_migrate_second_boot_preserves_env_byte_for_byte(tmp_path:
     assert env_path.read_bytes() == env_bytes_before
 
     config_after_first = config_path.read_bytes()
-    first_boot_backups = sorted(tmp_path.glob("config.yaml.bak-*"))
+    first_boot_backups = sorted((tmp_path / "backups" / "config").glob("config.yaml.pre-docker-migrate.*"))
 
     # ── Second boot (host reboot): version is current, must be a no-op. ──
     second = _run_migration(tmp_path)
@@ -255,4 +255,4 @@ def test_docker_config_migrate_second_boot_preserves_env_byte_for_byte(tmp_path:
     assert env_path.read_bytes() == env_bytes_before
     # config.yaml is untouched by the second boot, and no new backup is made.
     assert config_path.read_bytes() == config_after_first
-    assert sorted(tmp_path.glob("config.yaml.bak-*")) == first_boot_backups
+    assert sorted((tmp_path / "backups" / "config").glob("config.yaml.pre-docker-migrate.*")) == first_boot_backups

website/docs/integrations/providers.md

@@ -355,7 +355,7 @@ No configuration is needed — caching activates automatically when an xAI endpo
 
 xAI also ships a dedicated TTS endpoint (`/v1/tts`). Select **xAI TTS** in `hermes tools` → Voice & TTS, or see the [Voice & TTS](../user-guide/features/tts.md#text-to-speech) page for config.
 
-**Retired xAI model migration (May 15, 2026):** xAI is retiring `grok-4*`, `grok-3`, `grok-code-fast-1`, and `grok-imagine-image-pro` on 2026-05-15. `hermes doctor` and `hermes chat` startup both detect any config still pointing at a retired ref and print the recommended replacement. Use `hermes migrate xai` for a one-shot config rewrite — dry-run by default, add `--apply` to write changes (a timestamped `config.yaml.bak-pre-migrate-xai-*` backup is created automatically).
+**Retired xAI model migration (May 15, 2026):** xAI is retiring `grok-4*`, `grok-3`, `grok-code-fast-1`, and `grok-imagine-image-pro` on 2026-05-15. `hermes doctor` and `hermes chat` startup both detect any config still pointing at a retired ref and print the recommended replacement. Use `hermes migrate xai` for a one-shot config rewrite — dry-run by default, add `--apply` to write changes (a timestamped copy of the previous config lands in `backups/config/` first).
 
 ```bash
 hermes migrate xai          # preview replacements

website/docs/user-guide/configuration.md

@@ -182,6 +182,8 @@ updates:
 
 `pre_update_backup` is the single pre-update safety knob: `quick` (default) snapshots critical state files (pairing data, cron jobs, config, auth; files over 1 GiB are skipped) into `state-snapshots/`; `full` additionally zips all of `HERMES_HOME` into `backups/` and can add minutes on large homes; `off` disables both. Legacy booleans are honored (`true` → `full`, `false` → `off`).
 
+Point-in-time copies of `config.yaml` itself (taken before `hermes setup` rewrites it, before `hermes migrate` edits it, and when the file fails to parse) go to `backups/config/config.yaml.<reason>.<timestamp>`. Identical repeats are skipped and only the newest five per reason are kept, so they never pile up beside `config.yaml`.
+
 For git installs, Hermes auto-stashes dirty tracked files and untracked files before checking out the update branch or pulling. Interactive terminal updates prompt before restoring that stash. Non-interactive updates (desktop/chat app, gateway, or `--yes`) use `updates.non_interactive_local_changes`: `stash` restores local source edits after a successful pull, while `discard` drops the update-created stash after a successful pull. Use `discard` only on managed installs where local source edits are never meant to persist.
 
 Before that stash step, Hermes also restores tracked `package-lock.json` diffs left by npm install/build churn. Commit or manually stash intentional lockfile edits before updating.