From 36e894aa628ca569e171f6f76ebc5ef04cb6bf10 Mon Sep 17 00:00:00 2001
From: Safi <safishamsi98@gmail.com>
Date: Sat, 2 May 2026 14:25:26 +0100
Subject: [PATCH] v0.6.6: Windows skill bash rewrite, wiki fixes,
 rationale-node fix, hidden allowlist, --no-viz cluster-only

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 CHANGELOG.md              |  13 ++++
 graphify/__main__.py      |  26 ++++++--
 graphify/detect.py        | 121 ++++++++++++++++++++++++++++++++++++--
 graphify/export.py        |  44 ++++++++++++--
 graphify/extract.py       |  23 +++++++-
 graphify/skill-trae.md    |   2 +-
 graphify/skill-windows.md |  33 +++++++++--
 graphify/skill.md         |  27 +++++++--
 graphify/wiki.py          |  22 ++++++-
 pyproject.toml            |   2 +-
 10 files changed, 283 insertions(+), 30 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 2fb5f9d5..4939e65d 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,19 @@
 
 Full release notes with details on each version: [GitHub Releases](https://github.com/safishamsi/graphify/releases)
 
+## 0.6.6 (2026-05-02)
+
+- Fix: `skill-windows.md` rewritten from PowerShell to bash — Claude Code on Windows uses git-bash so PowerShell syntax (`$null`, `$LASTEXITCODE`, `Select-Object`, `& (Get-Content ...)`, `Remove-Item`) caused exit code 49 failures; now mirrors `skill.md` structure with `python` added as fallback after `python3` for Windows Conda (#39)
+- Fix: wiki `to_wiki()` now clears stale articles before regenerating, preventing orphan .md accumulation (#558)
+- Fix: `_safe_filename()` in wiki.py now strips Windows-reserved characters (`< > : " / \ | ? *`) and caps length at 200 chars (#594)
+- Fix: rationale-node leakage in cross-file INFERRED call resolution — rationale nodes now excluded from name lookup; edge direction (`calls`, `rationale_for`) preserved correctly at JSON export (#576)
+- Feat: `.graphifyinclude` hidden path allowlist — opt specific hidden dirs into traversal (e.g. `.hermes/plans/**/*.md`) (#583)
+- Feat: `--no-viz` flag wired in `cluster-only`; `GRAPHIFY_VIZ_NODE_LIMIT` env var overrides 5000-node HTML threshold (#565)
+- Fix: stray colon SyntaxError in `skill-trae.md` `--cluster-only` block (#603)
+- Docs: skill INFERRED confidence score guidance changed to discrete rubric (0.55/0.65/0.75/0.85/0.95) backed by calibration data (#546)
+- Docs: skill `--update` prune output clarified — splits no-drift vs drift cases (#544)
+- Docs: skill `--update` merge step now calls `save_manifest` to prevent deleted files reappearing (#545)
+
 ## 0.6.5 (2026-05-02)
 
 - Fix: Kotlin call-walker now accepts both `simple_identifier` and `identifier` node types — PyPI's `tree_sitter_kotlin` grammar uses `identifier` while older forks use `simple_identifier`, causing zero `calls` edges to be emitted (#659)
diff --git a/graphify/__main__.py b/graphify/__main__.py
index 94d1f69c..42d4429a 100644
--- a/graphify/__main__.py
+++ b/graphify/__main__.py
@@ -1008,6 +1008,7 @@ def main() -> None:
         print("    --force                 overwrite graph.json even if the rebuild has fewer nodes")
         print("                            (also: GRAPHIFY_FORCE=1 env var; use after refactors that delete code)")
         print("  cluster-only <path>     rerun clustering on an existing graph.json and regenerate report")
+        print("    --no-viz                skip graph.html generation (useful for >5000 node graphs / CI)")
         print("  query \"<question>\"       BFS traversal of graph.json for a question")
         print("    --dfs                   use depth-first instead of breadth-first")
         print("    --budget N              cap output at N tokens (default 2000)")
@@ -1385,6 +1386,7 @@ def main() -> None:
 
     elif cmd == "cluster-only":
         watch_path = Path(sys.argv[2]) if len(sys.argv) > 2 else Path(".")
+        no_viz = "--no-viz" in sys.argv
         graph_json = watch_path / "graphify-out" / "graph.json"
         if not graph_json.exists():
             print(f"error: no graph found at {graph_json} — run /graphify first", file=sys.stderr)
@@ -1414,11 +1416,25 @@ def main() -> None:
         out = watch_path / "graphify-out"
         (out / "GRAPH_REPORT.md").write_text(report, encoding="utf-8")
         to_json(G, communities, str(out / "graph.json"))
-        try:
-            to_html(G, communities, str(out / "graph.html"), community_labels=labels or None)
-        except ValueError as _viz_err:
-            print(f"[graphify] Skipped graph.html: {_viz_err}")
-        print(f"Done — {len(communities)} communities. GRAPH_REPORT.md, graph.json and graph.html updated.")
+
+        # Mirror watch.py pattern: gate to_html so core outputs (graph.json +
+        # GRAPH_REPORT.md) always land. Honor --no-viz explicitly; otherwise
+        # fall back to ValueError handling so an oversized graph doesn't crash
+        # the CLI mid-write and leave a stale graph.html on disk.
+        html_target = out / "graph.html"
+        if no_viz:
+            if html_target.exists():
+                html_target.unlink()
+            print(f"Done — {len(communities)} communities. GRAPH_REPORT.md and graph.json updated (--no-viz; graph.html removed).")
+        else:
+            try:
+                to_html(G, communities, str(html_target), community_labels=labels or None)
+                print(f"Done — {len(communities)} communities. GRAPH_REPORT.md, graph.json and graph.html updated.")
+            except ValueError as viz_err:
+                if html_target.exists():
+                    html_target.unlink()
+                print(f"Skipped graph.html: {viz_err}")
+                print(f"Done — {len(communities)} communities. GRAPH_REPORT.md and graph.json updated.")
 
     elif cmd == "update":
         force = os.environ.get("GRAPHIFY_FORCE", "").lower() in ("1", "true", "yes")
diff --git a/graphify/detect.py b/graphify/detect.py
index ba1595e6..419fc5c9 100644
--- a/graphify/detect.py
+++ b/graphify/detect.py
@@ -509,6 +509,115 @@ def _is_ignored(path: Path, root: Path, patterns: list[tuple[Path, str]]) -> boo
     return result
 
 
+def _load_graphifyinclude(root: Path) -> list[tuple[Path, str]]:
+    """Read .graphifyinclude allowlist patterns from root and ancestors.
+
+    Include patterns opt matching hidden files/dirs into traversal. Sensitive
+    files and hard-skipped noise directories are still excluded later.
+    Uses the same VCS-root ceiling logic as _load_graphifyignore.
+    """
+    root = root.resolve()
+    ceiling = _find_vcs_root(root) or root
+
+    dirs: list[Path] = []
+    current = root
+    while True:
+        dirs.append(current)
+        if current == ceiling:
+            break
+        current = current.parent
+    dirs.reverse()
+
+    patterns: list[tuple[Path, str]] = []
+    for d in dirs:
+        include_file = d / ".graphifyinclude"
+        if include_file.exists():
+            for raw in include_file.read_text(encoding="utf-8", errors="ignore").splitlines():
+                line = _parse_gitignore_line(raw)
+                if line:
+                    patterns.append((d, line))
+    return patterns
+
+
+def _is_included(path: Path, root: Path, patterns: list[tuple[Path, str]]) -> bool:
+    """Return True if path matches any .graphifyinclude allowlist pattern."""
+    if not patterns:
+        return False
+
+    def _matches(rel: str, p: str) -> bool:
+        parts = rel.split("/")
+        if fnmatch.fnmatch(rel, p):
+            return True
+        if fnmatch.fnmatch(path.name, p):
+            return True
+        for i, part in enumerate(parts):
+            if fnmatch.fnmatch(part, p):
+                return True
+            if fnmatch.fnmatch("/".join(parts[:i + 1]), p):
+                return True
+        return False
+
+    for anchor, pattern in patterns:
+        anchored = pattern.startswith("/")
+        p = pattern.strip("/")
+        if not p:
+            continue
+        if anchored:
+            try:
+                rel_anchor = str(path.relative_to(anchor)).replace(os.sep, "/")
+                if _matches(rel_anchor, p):
+                    return True
+            except ValueError:
+                pass
+        else:
+            try:
+                rel = str(path.relative_to(root)).replace(os.sep, "/")
+                if _matches(rel, p):
+                    return True
+            except ValueError:
+                pass
+            if anchor != root:
+                try:
+                    rel_anchor = str(path.relative_to(anchor)).replace(os.sep, "/")
+                    if _matches(rel_anchor, p):
+                        return True
+                except ValueError:
+                    pass
+    return False
+
+
+def _could_contain_included_path(path: Path, root: Path, patterns: list[tuple[Path, str]]) -> bool:
+    """Return True if a directory may contain files matched by .graphifyinclude."""
+    if not patterns:
+        return False
+
+    rels: list[str] = []
+    try:
+        rels.append(str(path.relative_to(root)).replace(os.sep, "/"))
+    except ValueError:
+        pass
+    for anchor, _ in patterns:
+        if anchor != root:
+            try:
+                rels.append(str(path.relative_to(anchor)).replace(os.sep, "/"))
+            except ValueError:
+                pass
+
+    for rel in rels:
+        rel = rel.strip("/")
+        if not rel:
+            return True
+        for _, pattern in patterns:
+            p = pattern.strip("/")
+            if not p:
+                continue
+            if p == rel or p.startswith(rel + "/"):
+                return True
+            if fnmatch.fnmatch(rel, p):
+                return True
+    return False
+
+
 def detect(root: Path, *, follow_symlinks: bool = False) -> dict:
     root = root.resolve()
     files: dict[FileType, list[str]] = {
@@ -522,6 +631,7 @@ def detect(root: Path, *, follow_symlinks: bool = False) -> dict:
 
     skipped_sensitive: list[str] = []
     ignore_patterns = _load_graphifyignore(root)
+    include_patterns = _load_graphifyinclude(root)
 
     # Always include graphify-out/memory/ - query results filed back into the graph
     memory_dir = root / "graphify-out" / "memory"
@@ -543,10 +653,12 @@ def detect(root: Path, *, follow_symlinks: bool = False) -> dict:
                     dirnames.clear()
                     continue
             if not in_memory_tree:
-                # Prune noise dirs in-place so os.walk never descends into them
+                # Prune noise dirs in-place so os.walk never descends into them.
+                # Hidden dirs are allowed through if they could contain an
+                # explicitly included path (.graphifyinclude allowlist).
                 dirnames[:] = [
                     d for d in dirnames
-                    if not d.startswith(".")
+                    if (not d.startswith(".") or _could_contain_included_path(dp / d, root, include_patterns))
                     and not _is_noise_dir(d)
                     and not _is_ignored(dp / d, root, ignore_patterns)
                 ]
@@ -565,8 +677,9 @@ def detect(root: Path, *, follow_symlinks: bool = False) -> dict:
         in_memory = memory_dir.exists() and str(p).startswith(str(memory_dir))
         if not in_memory:
             # Hidden files are already excluded via dir pruning above,
-            # but catch hidden files at the root level
-            if p.name.startswith("."):
+            # but catch hidden files at the root level. A .graphifyinclude
+            # entry can opt a specific hidden file back in.
+            if p.name.startswith(".") and not _is_included(p, root, include_patterns):
                 continue
             # Skip files inside our own converted/ dir (avoid re-processing sidecars)
             if str(p).startswith(str(converted_dir)):
diff --git a/graphify/export.py b/graphify/export.py
index 1d45e53a..14587add 100644
--- a/graphify/export.py
+++ b/graphify/export.py
@@ -25,6 +25,22 @@ COMMUNITY_COLORS = [
 MAX_NODES_FOR_VIZ = 5_000
 
 
+def _viz_node_limit() -> int:
+    """Return the effective viz node limit, honoring GRAPHIFY_VIZ_NODE_LIMIT env var.
+
+    Falls back to MAX_NODES_FOR_VIZ when the env var is unset, empty, or non-integer.
+    Set to 0 to disable HTML viz unconditionally (useful for CI runners).
+    """
+    import os
+    raw = os.environ.get("GRAPHIFY_VIZ_NODE_LIMIT")
+    if raw is None or not raw.strip():
+        return MAX_NODES_FOR_VIZ
+    try:
+        return int(raw)
+    except ValueError:
+        return MAX_NODES_FOR_VIZ
+
+
 def _html_styles() -> str:
     return """<style>
   * { box-sizing: border-box; margin: 0; padding: 0; }
@@ -357,6 +373,15 @@ def to_json(G: nx.Graph, communities: dict[int, list[str]], output_path: str, *,
         if "confidence_score" not in link:
             conf = link.get("confidence", "EXTRACTED")
             link["confidence_score"] = _CONFIDENCE_SCORE_DEFAULTS.get(conf, 1.0)
+        # Restore original edge direction. Undirected NetworkX storage may
+        # canonicalize endpoint order, flipping `calls` and other directional
+        # edges in graph.json. The build path stashes the true endpoints in
+        # _src/_tgt for exactly this purpose (#563).
+        true_src = link.pop("_src", None)
+        true_tgt = link.pop("_tgt", None)
+        if true_src is not None and true_tgt is not None:
+            link["source"] = true_src
+            link["target"] = true_tgt
     data["hyperedges"] = getattr(G, "graph", {}).get("hyperedges", [])
     with open(output_path, "w", encoding="utf-8") as f:  # nosec
         json.dump(data, f, indent=2)
@@ -421,10 +446,12 @@ def to_html(
     If member_counts is provided (aggregated community view), node sizes are
     based on community member counts rather than graph degree.
     """
-    if G.number_of_nodes() > MAX_NODES_FOR_VIZ:
+    limit = _viz_node_limit()
+    if G.number_of_nodes() > limit:
         raise ValueError(
-            f"Graph has {G.number_of_nodes()} nodes - too large for HTML viz. "
-            f"Use --no-viz or reduce input size."
+            f"Graph has {G.number_of_nodes()} nodes - too large for HTML viz "
+            f"(limit: {limit}). Use --no-viz, raise GRAPHIFY_VIZ_NODE_LIMIT, "
+            f"or reduce input size."
         )
 
     node_community = _node_community_map(communities)
@@ -461,14 +488,19 @@ def to_html(
             "degree": deg,
         })
 
-    # Build edges list
+    # Build edges list. Restore original edge direction from _src/_tgt
+    # (stashed by build.py for exactly this reason): undirected NetworkX
+    # canonicalizes endpoint order, which would otherwise flip the arrow
+    # for `calls` and `rationale_for` in the rendered graph (#563).
     vis_edges = []
     for u, v, data in G.edges(data=True):
         confidence = data.get("confidence", "EXTRACTED")
         relation = data.get("relation", "")
+        true_src = data.get("_src", u)
+        true_tgt = data.get("_tgt", v)
         vis_edges.append({
-            "from": u,
-            "to": v,
+            "from": true_src,
+            "to": true_tgt,
             "label": relation,
             "title": _html.escape(f"{relation} [{confidence}]"),
             "dashes": confidence != "EXTRACTED",
diff --git a/graphify/extract.py b/graphify/extract.py
index d45a97a7..1f7ec23d 100644
--- a/graphify/extract.py
+++ b/graphify/extract.py
@@ -2838,8 +2838,17 @@ def _resolve_cross_file_imports(
             stem = Path(src).stem
             label = node.get("label", "")
             nid = node.get("id", "")
-            # Only index real classes/functions (not file nodes, not method stubs)
-            if label and not label.endswith((")", ".py")) and "_" not in label[:1]:
+            # Index class-level entities only. Function/method labels end in "()"
+            # so are excluded by the `endswith(")")` filter; file nodes end in ".py";
+            # private/internal labels start with "_"; rationale nodes carry
+            # file_type=="rationale" and must never participate in cross-file
+            # import resolution (#563).
+            if (
+                label
+                and not label.endswith((")", ".py"))
+                and "_" not in label[:1]
+                and node.get("file_type") != "rationale"
+            ):
                 stem_to_entities.setdefault(stem, {})[label] = nid
 
     # Pass 2: for each file, find `from .X import A, B, C` and resolve
@@ -2850,12 +2859,15 @@ def _resolve_cross_file_imports(
         stem = _file_stem(path)
         str_path = str(path)
 
-        # Find all classes defined in this file (the importers)
+        # Find all classes defined in this file (the importers).
+        # Excludes rationale nodes whose labels happen not to end in ")" or ".py"
+        # but which must never be treated as importing entities (#563).
         local_classes = [
             n["id"] for n in file_result.get("nodes", [])
             if n.get("source_file") == str_path
             and not n["label"].endswith((")", ".py"))
             and n["id"] != _make_id(stem)  # exclude file-level node
+            and n.get("file_type") != "rationale"
         ]
         if not local_classes:
             continue
@@ -3578,8 +3590,13 @@ def extract(paths: list[Path], cache_root: Path | None = None) -> dict:
     # Build name → ALL matching node IDs so we can skip ambiguous common names
     # (e.g. "log", "execute", "find") that appear in multiple files — resolving
     # those inflates god_nodes ranking with spurious cross-file edges.
+    # Build label -> node_id index for cross-file call resolution.
+    # Skip rationale nodes (their labels are docstring text, not callable
+    # identifiers, and they were polluting matches for short names — #563).
     global_label_to_nids: dict[str, list[str]] = {}
     for n in all_nodes:
+        if n.get("file_type") == "rationale":
+            continue
         raw = n.get("label", "")
         normalised = raw.strip("()").lstrip(".")
         if normalised:
diff --git a/graphify/skill-trae.md b/graphify/skill-trae.md
index 2b5b401e..5f5597e6 100644
--- a/graphify/skill-trae.md
+++ b/graphify/skill-trae.md
@@ -826,7 +826,7 @@ G = json_graph.node_link_graph(data, edges='links')
 
 detection = {'total_files': 0, 'total_words': 99999, 'needs_graph': True, 'warning': None,
              'files': {'code': [], 'document': [], 'paper': []}}
-tokens = {'input': 0, 'output':': 0}
+tokens = {'input': 0, 'output': 0}
 
 communities = cluster(G)
 cohesion = score_all(G, communities)
diff --git a/graphify/skill-windows.md b/graphify/skill-windows.md
index 9984022e..cca70b93 100644
--- a/graphify/skill-windows.md
+++ b/graphify/skill-windows.md
@@ -279,10 +279,16 @@ If a file has YAML frontmatter (--- ... ---), copy source_url, captured_at, auth
 
 confidence_score is REQUIRED on every edge - never omit it, never use 0.5 as a default:
 - EXTRACTED edges: confidence_score = 1.0 always
-- INFERRED edges: reason about each edge individually.
-  Direct structural evidence (shared data structure, clear dependency): 0.8-0.9.
-  Reasonable inference with some uncertainty: 0.6-0.7.
-  Weak or speculative: 0.4-0.5. Most edges should be 0.6-0.9, not 0.5.
+- INFERRED edges: pick exactly ONE value from this set — never 0.5:
+    0.95  direct structural evidence (shared data structure, named cross-file reference).
+    0.85  strong inference (clear functional alignment, no direct symbol link).
+    0.75  reasonable inference (shared problem domain + similar shape, requires interpretation).
+    0.65  weak inference (thematically related, no shape evidence).
+    0.55  speculative but plausible (surface-level co-occurrence only).
+  Models follow discrete rubrics better than continuous ranges; the bimodal
+  distribution observed in production (>50% at 0.5, >40% at 0.85+) shows the
+  range guidance is being collapsed to a binary. If no value above fits, mark
+  the edge AMBIGUOUS rather than picking 0.4 or below.
 - AMBIGUOUS edges: 0.1-0.3
 
 Output exactly this JSON (no other text):
@@ -785,9 +791,28 @@ G_existing = json_graph.node_link_graph(existing_data, edges='links')
 new_extraction = json.loads(Path('.graphify_extract.json').read_text())
 G_new = build_from_json(new_extraction)
 
+# Prune nodes from deleted files
+incremental = json.loads(Path('.graphify_incremental.json').read_text())
+deleted = set(incremental.get('deleted_files', []))
+if deleted:
+    to_remove = [n for n, d in G_existing.nodes(data=True) if d.get('source_file') in deleted]
+    G_existing.remove_nodes_from(to_remove)
+    if to_remove:
+        print(f'Pruned {len(to_remove)} ghost node(s) from {len(deleted)} deleted file(s) — drift detected and corrected.')
+    else:
+        print(f'{len(deleted)} file(s) deleted since last run, but no ghost nodes were present in the graph — no drift.')
+
 # Merge: new nodes/edges into existing graph
 G_existing.update(G_new)
 print(f'Merged: {G_existing.number_of_nodes()} nodes, {G_existing.number_of_edges()} edges')
+
+# Save manifest with the CURRENT full file list so the next --update
+# diffs against today's filesystem state, not the prior --update's
+# baseline. Without this, deleted files get reported as ghosts again
+# on every subsequent --update until a full rebuild runs.
+from graphify.detect import save_manifest
+save_manifest(incremental['files'])
+print('[graphify update] Manifest saved.')
 " 
 ```
 
diff --git a/graphify/skill.md b/graphify/skill.md
index 10e4b701..c49cbfb3 100644
--- a/graphify/skill.md
+++ b/graphify/skill.md
@@ -337,10 +337,16 @@ If a file has YAML frontmatter (--- ... ---), copy source_url, captured_at, auth
 
 confidence_score is REQUIRED on every edge - never omit it, never use 0.5 as a default:
 - EXTRACTED edges: confidence_score = 1.0 always
-- INFERRED edges: reason about each edge individually.
-  Direct structural evidence (shared data structure, clear dependency): 0.8-0.9.
-  Reasonable inference with some uncertainty: 0.6-0.7.
-  Weak or speculative: 0.4-0.5. Most edges should be 0.6-0.9, not 0.5.
+- INFERRED edges: pick exactly ONE value from this set — never 0.5:
+    0.95  direct structural evidence (shared data structure, named cross-file reference).
+    0.85  strong inference (clear functional alignment, no direct symbol link).
+    0.75  reasonable inference (shared problem domain + similar shape, requires interpretation).
+    0.65  weak inference (thematically related, no shape evidence).
+    0.55  speculative but plausible (surface-level co-occurrence only).
+  Models follow discrete rubrics better than continuous ranges; the bimodal
+  distribution observed in production (>50% at 0.5, >40% at 0.85+) shows the
+  range guidance is being collapsed to a binary. If no value above fits, mark
+  the edge AMBIGUOUS rather than picking 0.4 or below.
 - AMBIGUOUS edges: 0.1-0.3
 
 Node ID format: lowercase, only `[a-z0-9_]`, no dots or slashes. Format: `{stem}_{entity}` where stem is the filename without extension and entity is the symbol name, both normalized (lowercase, non-alphanumeric chars replaced with `_`). Example: `src/auth/session.py` + `ValidateToken` → `session_validatetoken`. This must match the ID the AST extractor generates so cross-references between code and semantic nodes connect correctly. CRITICAL: never append chunk numbers, sequence numbers, or any suffix to an ID (no `_c1`, `_c2`, `_chunk2`, etc.). IDs must be deterministic from the label alone — the same entity must always produce the same ID regardless of which chunk processes it.
@@ -923,7 +929,10 @@ deleted = set(incremental.get('deleted_files', []))
 if deleted:
     to_remove = [n for n, d in G_existing.nodes(data=True) if d.get('source_file') in deleted]
     G_existing.remove_nodes_from(to_remove)
-    print(f'Pruned {len(to_remove)} ghost nodes from {len(deleted)} deleted file(s)')
+    if to_remove:
+        print(f'Pruned {len(to_remove)} ghost node(s) from {len(deleted)} deleted file(s) — drift detected and corrected.')
+    else:
+        print(f'{len(deleted)} file(s) deleted since last run, but no ghost nodes were present in the graph — no drift.')
 
 # Merge: new nodes/edges into existing graph
 G_existing.update(G_new)
@@ -939,6 +948,14 @@ merged_out = {
 }
 Path('graphify-out/.graphify_extract.json').write_text(json.dumps(merged_out))
 print(f'[graphify update] Merged extraction written ({len(merged_out[\"nodes\"])} nodes, {len(merged_out[\"edges\"])} edges)')
+
+# Save manifest with the CURRENT full file list so the next --update
+# diffs against today's filesystem state, not the prior --update's
+# baseline. Without this, deleted files get reported as ghosts again
+# on every subsequent --update until a full rebuild runs.
+from graphify.detect import save_manifest
+save_manifest(incremental['files'])
+print('[graphify update] Manifest saved.')
 " 
 ```
 
diff --git a/graphify/wiki.py b/graphify/wiki.py
index 6b2f5f95..8cc845f7 100644
--- a/graphify/wiki.py
+++ b/graphify/wiki.py
@@ -7,7 +7,18 @@ import networkx as nx
 
 
 def _safe_filename(name: str) -> str:
-    return name.replace("/", "-").replace(" ", "_").replace(":", "-")
+    """Make a label safe for use as a filename across platforms.
+
+    Substitutes characters that Windows reserves in filenames
+    (< > : " / \\ | ? *) and strips trailing dots/spaces, also reserved.
+    Falls back to 'unnamed' for empty results and caps length at 200
+    chars to stay well under common filesystem limits.
+    """
+    import re
+    s = name.replace("/", "-").replace(" ", "_").replace(":", "-")
+    s = re.sub(r'[<>:"/\\|?*]', '_', s)
+    s = s.strip('. ')
+    return s[:200] if s else 'unnamed'
 
 
 def _cross_community_links(G: nx.Graph, nodes: list[str], own_cid: int, labels: dict[int, str]) -> list[tuple[str, int]]:
@@ -185,6 +196,15 @@ def to_wiki(
     out = Path(output_dir)
     out.mkdir(parents=True, exist_ok=True)
 
+    # Clear stale .md files from previous runs to prevent orphan accumulation.
+    # Community labels are LLM-generated (per skill.md Step 5) and non-deterministic
+    # across runs — the same conceptual community may be named differently each time
+    # (e.g. "AutoAgent Skills" → "AutoAgent Methodology"), leaving the previous file
+    # as an orphan. Since to_wiki() owns wiki/ entirely (always writes the full set),
+    # it can safely clear .md files at the start of each call.
+    for old_article in out.glob("*.md"):
+        old_article.unlink()
+
     labels = community_labels or {cid: f"Community {cid}" for cid in communities}
     cohesion = cohesion or {}
     god_nodes_data = god_nodes_data or []
diff --git a/pyproject.toml b/pyproject.toml
index 2ffd2859..217c2d37 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "graphifyy"
-version = "0.6.5"
+version = "0.6.6"
 description = "AI coding assistant skill (Claude Code, Codex, OpenCode, Cursor, Gemini CLI, Aider, OpenClaw, Factory Droid, Trae, Hermes, Kiro, Google Antigravity) - turn any folder of code, docs, papers, images, or videos into a queryable knowledge graph"
 readme = "README.md"
 license = { file = "LICENSE" }