Complete librarian Stage 1: stale plans + MEMORY drift

Adds the last two deterministic Stage 1 checks and wires them into
scripts/librarian.sh:

- librarian_check_stale_plans.py: per docs/plans/*.md, uses
  `git log -1 --format=%cs` to get the last commit date and flags
  plans untouched in 60+ days. Plans should be either active or
  archived; long silence is drift.
- librarian_check_memory_drift.py: validates the user-memory dir
  (~/.claude/projects/<slug>/memory). Each MEMORY.md index line
  must point at a real file with frontmatter (name, description,
  type ∈ {user, feedback, project, reference}); no orphaned memory
  files. Memory dir is derived from repo root for portability.

Both run clean against current state.

Stage 1 declared complete. Promotable-archive and duplicate-prose
detection are moved to Stage 2 in the plan because both need LLM
judgment to avoid false positives — not pattern matching.

Next concrete step is the open Stage 1 finding from yesterday:
docs/performance.md at 914 lines, due for a split into sub-topic
notes.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

docs/plans/librarian.md

@@ -141,18 +141,35 @@ operator applies the patch.
   in AGENTS.md. Caught one real finding on first run:
   `docs/performance.md` at 914 lines — split into sub-topics deferred
   as a separate task.
+- ✅ Stage 1, piece 4: `scripts/librarian_check_stale_plans.py`. Uses
+  `git log -1 --format=%cs` per plan file; flags plans whose last
+  commit is older than 60 days. Clean on first run (all four plans
+  committed 2026-05-07).
+- ✅ Stage 1, piece 5: `scripts/librarian_check_memory_drift.py`.
+  Validates that every MEMORY.md index line points to a real file
+  with required frontmatter fields (`name`, `description`, `type` ∈
+  {user, feedback, project, reference}) and that no memory file is
+  orphaned from the index. Memory dir derived from repo root, so the
+  script is portable. Clean on first run.
 
-## Stage 1 Remaining Checks
+## Stage 1 Status: Complete
 
-- Stale plans (mtime + git-log staleness heuristic).
-- Promotable archive entries (deferred to Stage 2 — heuristic vs LLM
-  judgment is the open question).
-- MEMORY.md drift (index lines vs target file `description:` frontmatter).
-- Duplicate prose (high-overlap pairs across archive vs research).
+All deterministic checks land. Two remaining concepts intentionally
+moved out of Stage 1 because they require LLM judgment, not
+deterministic detection:
+
+- **Promotable archive entries** → Stage 2. "Durable conclusion"
+  detection is judgment, not pattern matching.
+- **Duplicate prose** → Stage 2. Shingled-overlap heuristics produce
+  too many false positives in this repo's mix of archive snapshots
+  and derived research notes; LLM should decide whether two passages
+  are the *same idea* vs the *same evidence*.
+
+Open Stage 1 finding to address: `docs/performance.md` at 914 lines.
 
 ## Next Concrete Step
 
-Stale plan check (`scripts/librarian_check_stale_plans.py`). Walks
-`docs/plans/*.md` and flags files whose mtime is older than N days
-AND whose path hasn't appeared in `git log` over the same window.
-This catches plans that drift out of mind without being archived.
+Address the open Stage 1 finding by splitting `docs/performance.md`
+into sub-topic notes under `docs/research/` (and dated archive
+entries where appropriate). Sketch the split as a 1-page sub-plan
+before doing the actual move so we don't shred a useful document.

scripts/librarian.sh

@@ -27,6 +27,18 @@ if ! uv run python scripts/librarian_check_oversize.py; then
 fi
 echo
 
+echo "→ Stale plans (no commit in 60 days)"
+if ! uv run python scripts/librarian_check_stale_plans.py; then
+    overall=1
+fi
+echo
+
+echo "→ MEMORY.md drift (index vs frontmatter)"
+if ! uv run python scripts/librarian_check_memory_drift.py; then
+    overall=1
+fi
+echo
+
 if [ "$overall" -eq 0 ]; then
     echo "All Stage 1 checks passed."
 else

scripts/librarian_check_memory_drift.py

@@ -0,0 +1,104 @@
+"""MEMORY.md drift check for librarian Stage 1.
+
+Validates the user-memory directory:
+- Each MEMORY.md index line points to a real memory file.
+- Each memory file has YAML frontmatter with required fields
+  (`name`, `description`, `type`) and a valid `type`.
+- Reports orphan memory files (exist on disk but missing from index).
+
+Memory dir is derived from the repo root, mirroring how Claude Code
+resolves project-scoped memory paths. If the dir doesn't exist (fresh
+checkout), the check exits 0 silently.
+
+Usage:
+    uv run python scripts/librarian_check_memory_drift.py
+"""
+
+from __future__ import annotations
+
+import re
+from pathlib import Path
+
+VALID_TYPES = {"user", "feedback", "project", "reference"}
+REQUIRED_FIELDS = ("name", "description", "type")
+
+INDEX_LINE = re.compile(r"^\s*-\s*\[([^\]]+)\]\(([^)]+)\)")
+FRONTMATTER = re.compile(r"\A---\s*\n(.*?)\n---\s*\n", re.DOTALL)
+
+
+def _repo_root() -> Path:
+    current = Path(__file__).resolve()
+    for parent in current.parents:
+        if (parent / "pyproject.toml").is_file():
+            return parent
+    raise RuntimeError("pyproject.toml을 찾을 수 없어 repository root를 판정할 수 없습니다.")
+
+
+def _memory_dir(root: Path) -> Path:
+    project_slug = str(root).replace("/", "-")
+    return Path.home() / ".claude" / "projects" / project_slug / "memory"
+
+
+def _parse_frontmatter(body: str) -> dict[str, str] | None:
+    m = FRONTMATTER.match(body)
+    if not m:
+        return None
+    fields: dict[str, str] = {}
+    for line in m.group(1).splitlines():
+        if ":" not in line:
+            continue
+        key, _, value = line.partition(":")
+        fields[key.strip()] = value.strip()
+    return fields
+
+
+def main() -> int:
+    root = _repo_root()
+    mem_dir = _memory_dir(root)
+    index_file = mem_dir / "MEMORY.md"
+    if not index_file.is_file():
+        print(f"MEMORY.md not found at {mem_dir}; skipping memory check.")
+        return 0
+
+    findings: list[str] = []
+    referenced: set[str] = set()
+
+    for raw in index_file.read_text(encoding="utf-8").splitlines():
+        m = INDEX_LINE.match(raw)
+        if not m:
+            continue
+        target = m.group(2).strip()
+        referenced.add(target)
+        target_path = mem_dir / target
+        if not target_path.is_file():
+            findings.append(f"index points to missing file: {target}")
+            continue
+        body = target_path.read_text(encoding="utf-8")
+        fields = _parse_frontmatter(body)
+        if fields is None:
+            findings.append(f"{target}: missing or malformed frontmatter")
+            continue
+        for key in REQUIRED_FIELDS:
+            if key not in fields:
+                findings.append(f"{target}: frontmatter missing `{key}`")
+        if "type" in fields and fields["type"] not in VALID_TYPES:
+            findings.append(
+                f"{target}: invalid type `{fields['type']}` (allowed: {sorted(VALID_TYPES)})"
+            )
+
+    on_disk = {p.name for p in mem_dir.glob("*.md") if p.name != "MEMORY.md"}
+    for orphan in sorted(on_disk - referenced):
+        findings.append(f"orphan memory file (not in MEMORY.md index): {orphan}")
+
+    if not findings:
+        print("MEMORY.md and memory files consistent.")
+        return 0
+
+    print("MEMORY.md drift findings:")
+    for entry in findings:
+        print(f"  {entry}")
+    return 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/librarian_check_stale_plans.py

@@ -0,0 +1,77 @@
+"""Stale plan check for librarian Stage 1.
+
+Walks docs/plans/*.md (top-level only — archive subdir is separate)
+and flags plans whose last git commit is older than STALE_DAYS.
+Plans should be either actively worked on or archived (moved to
+docs/plans/archive/) — long-untouched files are usually drift.
+
+Usage:
+    uv run python scripts/librarian_check_stale_plans.py
+"""
+
+from __future__ import annotations
+
+import subprocess
+from datetime import datetime, timedelta
+from pathlib import Path
+
+STALE_DAYS = 60
+
+
+def _repo_root() -> Path:
+    current = Path(__file__).resolve()
+    for parent in current.parents:
+        if (parent / "pyproject.toml").is_file():
+            return parent
+    raise RuntimeError("pyproject.toml을 찾을 수 없어 repository root를 판정할 수 없습니다.")
+
+
+def _last_commit_date(root: Path, rel: Path) -> str | None:
+    result = subprocess.run(
+        ["git", "log", "-1", "--format=%cs", "--", str(rel)],
+        capture_output=True,
+        text=True,
+        cwd=root,
+        check=False,
+    )
+    out = result.stdout.strip()
+    return out if out else None
+
+
+def main() -> int:
+    root = _repo_root()
+    plans_dir = root / "docs" / "plans"
+    if not plans_dir.is_dir():
+        print("docs/plans not found; skipping stale-plan check.")
+        return 0
+
+    cutoff = (datetime.now() - timedelta(days=STALE_DAYS)).date()
+
+    stale: list[tuple[Path, str]] = []
+    for path in sorted(plans_dir.glob("*.md")):
+        rel = path.relative_to(root)
+        last = _last_commit_date(root, rel)
+        if last is None:
+            stale.append((rel, "untracked"))
+            continue
+        try:
+            last_date = datetime.strptime(last, "%Y-%m-%d").date()
+        except ValueError:
+            continue
+        if last_date < cutoff:
+            stale.append((rel, last))
+
+    if not stale:
+        print(f"No stale plans (all touched within {STALE_DAYS} days).")
+        return 0
+
+    print(f"Stale plan candidates (last commit > {STALE_DAYS} days ago):")
+    for rel, last in stale:
+        print(f"  {last}  {rel}")
+    print()
+    print("Action: archive (mv to docs/plans/archive/) or update if still active.")
+    return 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())