sourcecode 1.30.17__py3-none-any.whl → 1.30.19__py3-none-any.whl

sourcecode/__init__.py

@@ -1,3 +1,3 @@
 """sourcecode — Deterministic codebase context maps for AI coding agents."""
 
-__version__ = "1.30.17"
+__version__ = "1.30.19"

sourcecode/cli.py

@@ -1648,6 +1648,12 @@ def prepare_context_cmd(
         "--format",
         help="Output format: json (default) | github-comment (Markdown PR comment for review-pr task)",
     ),
+    debug_perf: bool = typer.Option(
+        False,
+        "--debug-perf",
+        help="Emit per-phase timing to stderr (git scan ms, symptom scoring ms, total ms)",
+        hidden=True,
+    ),
 ) -> None:
     """Task-specific context for AI coding agents.
 
@@ -1714,9 +1720,21 @@ def prepare_context_cmd(
         raise typer.Exit()
 
     from dataclasses import asdict
+    import time as _time
 
     builder = TaskContextBuilder(target)
+    _t0 = _time.perf_counter()
     output = builder.build(task, since=since, symptom=symptom)
+    _t_total = (_time.perf_counter() - _t0) * 1000
+
+    if debug_perf:
+        _perf = getattr(output, "_perf_ms", {}) or {}
+        typer.echo(
+            f"[debug-perf] total={_t_total:.0f}ms"
+            + (f"  git_scan={_perf.get('git_scan_ms', '?')}ms" if "git_scan_ms" in _perf else "")
+            + (f"  symptom={_perf.get('symptom_ms', '?')}ms" if "symptom_ms" in _perf else ""),
+            err=True,
+        )
 
     # Task-specific content-filter: each task emphasizes different output fields.
     # Fields marked False are suppressed from this task's output to reduce noise.
@@ -1918,6 +1936,8 @@ def prepare_context_cmd(
         out["related_notes"] = output.related_notes
     if output.symptom_note:
         out["symptom_note"] = output.symptom_note
+    if output.symptom_explain:
+        out["symptom_explain"] = output.symptom_explain
     if llm_prompt:
         out["llm_prompt"] = builder.render_prompt(output)
 
@@ -2012,12 +2032,33 @@ def repo_ir_cmd(
         "--include-tests",
         help="Include test files in analysis (excluded by default)",
     ),
+    max_nodes: Optional[int] = typer.Option(
+        None,
+        "--max-nodes",
+        help="Limit graph.nodes to top N by impact score (reduces output size)",
+    ),
+    max_edges: Optional[int] = typer.Option(
+        None,
+        "--max-edges",
+        help="Limit graph.edges to N (priority: edges between kept nodes)",
+    ),
+    summary_only: bool = typer.Option(
+        False,
+        "--summary-only",
+        help="Omit full graph.nodes/edges; keep analysis summary, impact, and change_set (<300KB typical)",
+    ),
 ) -> None:
     """Deterministic symbol-level IR for Java repositories.
 
     \b
     Extracts symbols, relations, Spring roles, and (with --since) symbol-level diffs.
-    Output is JSON: symbols[], relations[], changed_symbols[], spring_summary, graph_metadata.
+    Output is JSON: graph{nodes,edges}, analysis, impact, subsystems, change_set.
+
+    \b
+    Size control:
+      --summary-only          Omit full graph; keep analysis + impact (smallest output)
+      --max-nodes N           Keep top N nodes by score
+      --max-edges N           Keep top N edges (priority: both endpoints kept)
 
     \b
     Examples:
@@ -2025,11 +2066,13 @@ def repo_ir_cmd(
       sourcecode repo-ir /path/to/repo --since HEAD~1
       sourcecode repo-ir --files src/main/java/UserService.java
       sourcecode repo-ir --since main --output ir.json
+      sourcecode repo-ir --since HEAD~3 --summary-only --output ir-small.json
+      sourcecode repo-ir --max-nodes 200 --max-edges 500
     """
     import json as _json
     import sys as _sys
 
-    from sourcecode.repository_ir import build_repo_ir, find_java_files
+    from sourcecode.repository_ir import apply_ir_size_limits, build_repo_ir, find_java_files
 
     root = path.resolve()
     if not root.is_dir():
@@ -2063,11 +2106,24 @@ def repo_ir_cmd(
         return
 
     ir = build_repo_ir(file_list, root, since=since)
+    ir = apply_ir_size_limits(
+        ir,
+        max_nodes=max_nodes,
+        max_edges=max_edges,
+        summary_only=summary_only,
+    )
     output = _json.dumps(ir, indent=2, ensure_ascii=False)
 
     if output_path:
         output_path.write_text(output, encoding="utf-8")
-        typer.echo(f"IR written to {output_path}", err=True)
+        n_nodes = len((ir.get("graph") or {}).get("nodes") or [])
+        n_edges = len((ir.get("graph") or {}).get("edges") or [])
+        size_kb = len(output.encode("utf-8")) // 1024
+        typer.echo(
+            f"IR written to {output_path} "
+            f"({size_kb}KB, {n_nodes} nodes, {n_edges} edges)",
+            err=True,
+        )
     else:
         _sys.stdout.write(output)
         _sys.stdout.write("\n")

sourcecode/contract_pipeline.py

@@ -623,6 +623,8 @@ def _find_symbol_files(
             cwd=str(root),
             capture_output=True,
             text=True,
+            encoding="utf-8",
+            errors="replace",
             timeout=20,
         )
         for line in result.stdout.splitlines():

sourcecode/git_analyzer.py

@@ -52,8 +52,15 @@ _HOTSPOT_AUX_DIRS: frozenset[str] = frozenset({
     "sandbox", "sandboxes",
     "ci", "translations", "locales", "locale", "i18n", "l10n",
     ".planning",
+    # vendor / build dirs: high path count, zero operational signal
+    "vendor", "node_modules", "dist", "target", "build",
+    ".gradle", ".mvn", "generated-sources", "generated-resources",
 })
 
+# When git diff returns more changed files than this, degrade gracefully:
+# truncate hotspot analysis depth and emit a structured warning.
+_CHANGED_FILES_DEGRADATION_THRESHOLD = 200
+
 
 def _run_git(args: list[str], cwd: Path, timeout: int = 15) -> tuple[str, int]:
     result = subprocess.run(

sourcecode/prepare_context.py

@@ -333,6 +333,7 @@ class TaskOutput:
     symptom: Optional[str] = None                                  # fix-bug only
     related_notes: list[dict] = field(default_factory=list)        # fix-bug + symptom only
     symptom_note: Optional[str] = None                             # fix-bug: cross-layer synonym note
+    symptom_explain: Optional[dict] = None                         # fix-bug: structured evidence breakdown
     # delta-specific impact fields
     impact_summary: Optional[str] = None
     affected_modules: list[str] = field(default_factory=list)
@@ -561,17 +562,23 @@ _ARTIFACT_CHANGE_EFFECT: dict[str, str] = {
 # Maps frontend symptom keywords → backend terms likely to contain the root cause.
 # Used to boost service/interceptor files when the symptom is UI-only.
 _FRONTEND_SYMPTOM_MAP: dict[str, list[str]] = {
-    "spinner":  ["loading", "setloading", "finalize", "httpinterceptor", "interceptor", "service"],
-    "loading":  ["loading", "setloading", "finalize", "httpinterceptor", "interceptor", "service"],
-    "login":    ["authcontroller", "securityconfig", "filterconfig", "jwtfilter", "auth", "authentication"],
-    "logout":   ["authcontroller", "securityconfig", "jwtfilter", "auth", "session"],
-    "dropdown": ["getmapping", "findall", "obtenertodos", "listall", "findby"],
-    "modal":    ["controller", "getmapping", "findby", "search"],
-    "popup":    ["controller", "getmapping", "findby", "search"],
-    "table":    ["paginated", "findby", "search", "getmapping", "listall"],
-    "grid":     ["paginated", "findby", "search", "getmapping"],
-    "button":   ["postmapping", "putmapping", "deletemapping", "controller", "service"],
-    "form":     ["postmapping", "putmapping", "controller", "service", "dto"],
+    "spinner":    ["loading", "setloading", "finalize", "httpinterceptor", "interceptor", "service"],
+    "loading":    ["loading", "setloading", "finalize", "httpinterceptor", "interceptor", "service"],
+    "login":      ["authcontroller", "securityconfig", "filterconfig", "jwtfilter", "auth", "authentication"],
+    "logout":     ["authcontroller", "securityconfig", "jwtfilter", "auth", "session"],
+    "dropdown":   ["getmapping", "findall", "obtenertodos", "listall", "findby"],
+    "modal":      ["controller", "getmapping", "findby", "search"],
+    "popup":      ["controller", "getmapping", "findby", "search"],
+    "table":      ["paginated", "findby", "search", "getmapping", "listall"],
+    "grid":       ["paginated", "findby", "search", "getmapping"],
+    "button":     ["postmapping", "putmapping", "deletemapping", "controller", "service"],
+    "form":       ["postmapping", "putmapping", "controller", "service", "dto"],
+    # session-related symptoms (Spanish + English)
+    "sesion":     ["httpsession", "sessionmanager", "sessionservice", "sessionrepository", "sessionfactory", "authentication"],
+    "sesiones":   ["httpsession", "sessionmanager", "sessionservice", "sessionrepository", "sessionfactory", "authentication"],
+    "session":    ["httpsession", "sessionmanager", "sessionservice", "sessionrepository", "sessionfactory", "authentication"],
+    # worker/assignment domain terms (common in RRHH/HR systems)
+    "trabajador": ["trabajador", "empleado", "worker", "asignacion", "trabajadordao", "trabajadorservice"],
 }
 
 
@@ -812,7 +819,7 @@ class TaskContextBuilder:
                         if not self._is_source(_p) or self._is_test(_p):
                             continue
                         try:
-                            _loc = len((self.root / _p).read_text(errors="replace").splitlines())
+                            _loc = len((self.root / _p).read_text(encoding="utf-8", errors="replace").splitlines())
                             if _loc > 200:
                                 _large.append((_loc, _p))
                         except OSError:
@@ -826,12 +833,14 @@ class TaskContextBuilder:
         # ── 5b. Git signals for ranking ────────────────────────────────────
         git_hotspots: dict[str, int] = {}
         uncommitted_files: set[str] = set()
+        _recent_commits_for_symptom: list = []
         try:
             from sourcecode.git_analyzer import GitAnalyzer
             _gc = GitAnalyzer().analyze(self.root, depth=30, days=90)
             _bad = {"no_git_repo", "git_not_found", "git_timeout"}
             if _gc and not (_bad & set(_gc.limitations)):
                 git_hotspots = {h.file: h.commit_count for h in _gc.change_hotspots}
+                _recent_commits_for_symptom = list(_gc.recent_commits)
                 if _gc.uncommitted_changes:
                     _uc = _gc.uncommitted_changes
                     uncommitted_files = set(_uc.staged) | set(_uc.unstaged)
@@ -1199,6 +1208,7 @@ class TaskContextBuilder:
         symptom_keywords: list[str] = []
         related_notes: list[dict] = []
         symptom_note: Optional[str] = None
+        symptom_explain: Optional[dict] = None
         if task_name == "fix-bug" and symptom:
             import re as _re
             _camel_expanded = _re.sub(r'([a-z])([A-Z])', r'\1 \2', symptom)
@@ -1208,19 +1218,68 @@ class TaskContextBuilder:
                 if len(w) > 2
             ]
             if symptom_keywords:
-                # Surface code notes whose text contains any keyword
+                # Pre-compile combined keyword pattern for fast content scanning
+                _kw_re = _re.compile(
+                    "|".join(_re.escape(kw) for kw in symptom_keywords),
+                    _re.IGNORECASE,
+                )
+
+                # Structured evidence collectors for symptom_explain
+                _sx_direct_path: list[str] = []
+                _sx_content: list[str] = []
+                _sx_commits: list[dict] = []
+                _sx_synonyms: list[str] = []
+                _sx_boosts: list[dict] = []
+
+                # Pass 1: surface code notes whose text contains any keyword
+                _note_matched_paths: dict[str, int] = {}  # path → count of matching notes
                 for _n in cn_notes_for_ranking:
                     _text = (getattr(_n, "text", "") or "").lower()
-                    if any(kw in _text for kw in symptom_keywords):
+                    if _kw_re.search(_text):
+                        _np = getattr(_n, "path", "")
                         related_notes.append({
                             "kind": getattr(_n, "kind", ""),
-                            "path": getattr(_n, "path", ""),
+                            "path": _np,
                             "line": getattr(_n, "line", None),
                             "text": getattr(_n, "text", ""),
                         })
-                # Secondary pass: inject files whose path matches symptom keywords
-                # but weren't in the candidate pool (no structural/git signals).
+                        _note_matched_paths[_np] = _note_matched_paths.get(_np, 0) + 1
+
+                # Pass 2: build commit message index — cap at 60 most-recent commits.
+                # Files touched in commits whose message matches a symptom keyword get
+                # a strong recency signal. Primary signal for domain keywords ("sesiones")
+                # that appear in commit messages but not file paths.
+                _commit_file_hits: dict[str, int] = {}  # path → n matching commits
+                _commits_scanned = _recent_commits_for_symptom[:60]
+                for _cr in _commits_scanned:
+                    _msg_lower = (_cr.message or "").lower()
+                    if _kw_re.search(_msg_lower):
+                        for _cf in (_cr.files_changed or []):
+                            _cf_norm = _cf.replace("\\", "/")
+                            _commit_file_hits[_cf_norm] = _commit_file_hits.get(_cf_norm, 0) + 1
+                        _sx_commits.append({
+                            "message": (_cr.message or "")[:80],
+                            "files": list((_cr.files_changed or [])[:5]),
+                        })
+
+                # Pass 3: inject files from commit index not yet in candidate pool
                 _existing_paths = {rf.path for rf in relevant_files}
+                for _cp, _nhits in _commit_file_hits.items():
+                    if _cp in _existing_paths:
+                        continue
+                    if Path(_cp).suffix.lower() not in _ALL_EXTENSIONS:
+                        continue
+                    _ci_score = round(min(0.5 + 0.15 * _nhits, 0.85), 2)
+                    relevant_files.append(RelevantFile(
+                        path=_cp,
+                        role="symptom_match",
+                        score=_ci_score,
+                        reason=f"commit message matches symptom ({_nhits} commit{'s' if _nhits > 1 else ''})",
+                        why=f"symptom commit-index: {', '.join(symptom_keywords)}",
+                    ))
+                    _existing_paths.add(_cp)
+
+                # Pass 4: inject files whose path matches symptom keywords
                 for _p in all_paths:
                     if _p in _existing_paths:
                         continue
@@ -1241,72 +1300,134 @@ class TaskContextBuilder:
                         why=f"symptom injection: {', '.join(_matching_kws)}",
                     ))
                     _existing_paths.add(_p)
+                    _sx_direct_path.append(_p)
 
-                # Re-rank all relevant_files: boost files whose path matches keywords
-                def _symptom_score(rf: "RelevantFile") -> float:
-                    path_lower = rf.path.lower()
-                    return rf.score + 0.2 * sum(1.0 for kw in symptom_keywords if kw in path_lower)
-                relevant_files = sorted(relevant_files, key=lambda rf: -_symptom_score(rf))
-
-                # Content scan boost: read file body for symptom keywords
+                # Pass 5+6 combined: single file read per candidate.
+                # Limit content scan to top 80 candidates by current score to bound I/O.
                 _src_exts = frozenset({".java", ".py", ".ts", ".js", ".kt", ".go"})
-                _content_boosted: list[RelevantFile] = []
-                for _rf in relevant_files:
+                _frontend_kws = [kw for kw in symptom_keywords if kw in _FRONTEND_SYMPTOM_MAP]
+                _backend_terms_set: list[str] = []
+                if _frontend_kws:
+                    _bt: list[str] = []
+                    for _fkw in _frontend_kws:
+                        _bt.extend(_FRONTEND_SYMPTOM_MAP[_fkw])
+                    _backend_terms_set = list(dict.fromkeys(_bt))
+
+                # Sort before content scan so top candidates get read first
+                relevant_files = sorted(relevant_files, key=lambda rf: -rf.score)
+                _CONTENT_SCAN_LIMIT = 80
+                _scan_candidates = relevant_files[:_CONTENT_SCAN_LIMIT]
+                _no_scan_candidates = relevant_files[_CONTENT_SCAN_LIMIT:]
+
+                _boosted: list[RelevantFile] = []
+                for _rf in _scan_candidates:
                     _extra = 0.0
+                    _extra_syn = 0.0
+                    _reasons: list[str] = []
+                    _p_lower = _rf.path.lower()
+
+                    # Commit message boost: +0.25/commit, cap +0.40
+                    _c_hits = _commit_file_hits.get(_rf.path, 0)
+                    if _c_hits:
+                        _cb = min(0.40, _c_hits * 0.25)
+                        _extra += _cb
+                        _reasons.append(f"commit-msg symptom ×{_c_hits} (+{_cb:.2f})")
+                        _sx_boosts.append({"type": "commit_message", "value": round(_cb, 3), "evidence": _rf.path})
+
+                    # Code note boost: +0.20/note, cap +0.30
+                    _n_hits = _note_matched_paths.get(_rf.path, 0)
+                    if _n_hits:
+                        _nb = min(0.30, _n_hits * 0.20)
+                        _extra += _nb
+                        _reasons.append(f"note-match symptom ×{_n_hits} (+{_nb:.2f})")
+                        _sx_boosts.append({"type": "code_note", "value": round(_nb, 3), "evidence": _rf.path})
+
+                    # Path keyword boost for pre-existing candidates
+                    _path_kws = [kw for kw in symptom_keywords if kw in _p_lower]
+                    if _path_kws and _rf.role != "symptom_match":
+                        _pb = 0.20 * len(_path_kws)
+                        _extra += _pb
+                        _reasons.append(f"path-kw symptom ({', '.join(_path_kws)}) (+{_pb:.2f})")
+                        _sx_boosts.append({"type": "path_match", "value": round(_pb, 3), "evidence": _rf.path})
+
+                    # Single file read — covers both content scan and synonym scan
+                    _body_lower = ""
                     if Path(_rf.path).suffix.lower() in _src_exts:
                         try:
-                            _lines = (self.root / _rf.path).read_text(
+                            _body_lower = (self.root / _rf.path).read_text(
                                 encoding="utf-8", errors="replace"
-                            ).splitlines()[:300]
-                            _body = "\n".join(_lines).lower()
-                            _hits = sum(_body.count(kw) for kw in symptom_keywords)
-                            _extra = min(0.30, _hits * 0.02)
+                            )[:12000].lower()  # ~300 lines avg
                         except OSError:
                             pass
-                    _content_boosted.append(RelevantFile(
+
+                    # Content scan boost: +0.05/hit, cap +0.50
+                    if _body_lower:
+                        _hits = len(_kw_re.findall(_body_lower))
+                        _content_b = min(0.50, _hits * 0.05)
+                        if _content_b > 0:
+                            _extra += _content_b
+                            _reasons.append(f"content-match symptom ×{_hits} (+{_content_b:.2f})")
+                            _sx_boosts.append({"type": "content_match", "value": round(_content_b, 3), "evidence": _rf.path})
+                            _sx_content.append(_rf.path)
+
+                    # Synonym scan (Pass 6): only apply when file has prior non-synonym
+                    # evidence (commit hit, note, path, or content) — prevents boosting
+                    # arbitrary interceptors/configs with no other signal.
+                    if _backend_terms_set and _body_lower:
+                        _prior_boost = _extra  # boost accumulated before synonym
+                        if _prior_boost >= 0.10:  # min threshold: must have real prior signal
+                            _hits_syn = sum(_body_lower.count(t) for t in _backend_terms_set)
+                            _extra_syn = min(0.20, _hits_syn * 0.02)
+                            if _extra_syn > 0:
+                                _sx_synonyms.append(_rf.path)
+                                _sx_boosts.append({"type": "synonym_match", "value": round(_extra_syn, 3), "evidence": _rf.path})
+
+                    _total_extra = _extra + _extra_syn
+                    _new_reason = _rf.reason
+                    if _reasons:
+                        _syn_suffix = f", synonym-match backend (+{_extra_syn:.2f})" if _extra_syn > 0 else ""
+                        _new_reason = _rf.reason + ", " + ", ".join(_reasons) + _syn_suffix
+                    elif _extra_syn > 0:
+                        _new_reason = _rf.reason + f", synonym-match backend (+{_extra_syn:.2f})"
+
+                    _boosted.append(RelevantFile(
                         path=_rf.path,
                         role=_rf.role,
-                        score=round(min(_rf.score + _extra, 1.0), 2),
-                        reason=_rf.reason + (f", content-match symptom (+{_extra:.2f})" if _extra > 0 else ""),
+                        score=round(min(_rf.score + _total_extra, 1.0), 2),
+                        reason=_new_reason,
                         why=_rf.why,
                     ))
-                relevant_files = sorted(_content_boosted, key=lambda rf: -rf.score)
 
-                # Cross-layer synonym boost: frontend keywords → backend equivalents
-                _synonym_note: Optional[str] = None
-                _frontend_kws = [kw for kw in symptom_keywords if kw in _FRONTEND_SYMPTOM_MAP]
-                if _frontend_kws:
-                    _backend_terms: list[str] = []
-                    for _fkw in _frontend_kws:
-                        _backend_terms.extend(_FRONTEND_SYMPTOM_MAP[_fkw])
-                    _backend_terms_set = list(dict.fromkeys(_backend_terms))  # dedup, preserve order
-                    _synonym_boosted: list[RelevantFile] = []
-                    for _rf in relevant_files:
-                        _extra_syn = 0.0
-                        if Path(_rf.path).suffix.lower() in _src_exts:
-                            try:
-                                _lines_syn = (self.root / _rf.path).read_text(
-                                    encoding="utf-8", errors="replace"
-                                ).splitlines()[:300]
-                                _body_syn = "\n".join(_lines_syn).lower()
-                                _hits_syn = sum(_body_syn.count(t) for t in _backend_terms_set)
-                                _extra_syn = min(0.20, _hits_syn * 0.02)
-                            except OSError:
-                                pass
-                        _synonym_boosted.append(RelevantFile(
-                            path=_rf.path,
-                            role=_rf.role,
-                            score=round(min(_rf.score + _extra_syn, 1.0), 2),
-                            reason=_rf.reason + (f", synonym-match backend (+{_extra_syn:.2f})" if _extra_syn > 0 else ""),
-                            why=_rf.why,
-                        ))
-                    relevant_files = sorted(_synonym_boosted, key=lambda rf: -rf.score)
-                    _synonym_note = (
+                relevant_files = sorted(_boosted + _no_scan_candidates, key=lambda rf: -rf.score)
+
+                # Synonym note (only when synonyms actually fired)
+                if _frontend_kws and _sx_synonyms:
+                    symptom_note = (
                         f"Frontend concept detected ({', '.join(_frontend_kws)}). "
-                        "Backend service-layer and interceptor files boosted by symptom keyword match "
+                        "Backend service-layer files boosted by synonym match "
                         "[INFERRED (LOW CONFIDENCE) — pattern heuristic, not structural proof]."
                     )
-                    symptom_note = _synonym_note
+
+                # Confidence: based on richest signal type present
+                if _commit_file_hits:
+                    _sx_confidence = "HIGH"
+                elif _sx_direct_path or _sx_content:
+                    _sx_confidence = "MEDIUM"
+                else:
+                    _sx_confidence = "LOW"
+
+                symptom_explain = {
+                    "keywords": symptom_keywords,
+                    "confidence": _sx_confidence,
+                    "direct_path_matches": _sx_direct_path[:10],
+                    "content_matches": _sx_content[:10],
+                    "commit_matches": _sx_commits[:10],
+                    "synonym_matches": _sx_synonyms[:10],
+                    "boosts": _sx_boosts[:30],
+                    "final_boost": round(
+                        sum(b["value"] for b in _sx_boosts), 3
+                    ),
+                }
 
         # ── 7. Test gaps (generate-tests only) ────────────────────────────
         test_gaps: list[str] = []
@@ -1405,6 +1526,7 @@ class TaskContextBuilder:
             symptom=symptom if task_name == "fix-bug" and symptom else None,
             related_notes=related_notes,
             symptom_note=symptom_note,
+            symptom_explain=symptom_explain if task_name == "fix-bug" else None,
             impact_summary=_delta_impact_summary,
             affected_modules=_delta_affected_modules,
             risk_areas=_delta_risk_areas,

sourcecode/repository_ir.py

@@ -997,6 +997,8 @@ def _get_git_old_content(git_root: Path, rel_path: str, since: str) -> Optional[
             cwd=str(git_root),
             capture_output=True,
             text=True,
+            encoding="utf-8",
+            errors="replace",
             timeout=5,
         )
         if result.returncode == 0:
@@ -1542,12 +1544,99 @@ def build_repo_ir(
     return _assemble(all_symbols, unique_relations, all_changed, spring_summary, all_route_diffs_sorted)
 
 
+# ---------------------------------------------------------------------------
+# Output size limits
+# ---------------------------------------------------------------------------
+
+# Vendor/generated dirs to skip when finding Java files and in git analysis.
+_VENDOR_DIRS: frozenset[str] = frozenset({
+    "vendor", "node_modules", "dist", "target", "build",
+    ".gradle", ".mvn", "generated", "generated-sources",
+    "generated-resources",
+})
+
+
+def apply_ir_size_limits(
+    ir: dict,
+    *,
+    max_nodes: Optional[int] = None,
+    max_edges: Optional[int] = None,
+    summary_only: bool = False,
+) -> dict:
+    """Apply size limits to a repo-ir output dict. Non-destructive: returns new dict.
+
+    Node ordering: top-ranked (by impact score) nodes are kept first.
+    Edge priority: edges connecting two kept nodes over cross-boundary edges.
+    """
+    if not max_nodes and not max_edges and not summary_only:
+        return ir
+
+    out = dict(ir)
+    graph = ir.get("graph") or {}
+    nodes: list[dict] = list(graph.get("nodes") or [])
+    edges: list[dict] = list(graph.get("edges") or [])
+    ranked: list[dict] = list((ir.get("impact") or {}).get("ranked_nodes") or [])
+    analysis: dict = ir.get("analysis") or {}
+
+    if summary_only:
+        n_nodes, n_edges = len(nodes), len(edges)
+        out["graph"] = {
+            "nodes": [],
+            "edges": [],
+            "_omitted": (
+                f"{n_nodes} nodes and {n_edges} edges omitted — "
+                "remove --summary-only to restore full graph"
+            ),
+        }
+        out["reverse_graph"] = {}
+        out["impact"] = {
+            "global_score": (ir.get("impact") or {}).get("global_score", 0),
+            "ranked_nodes": ranked[:20],
+        }
+        out["analysis"] = {
+            "changed_entities": analysis.get("changed_entities") or [],
+            "impacted_entities": (analysis.get("impacted_entities") or [])[:20],
+            "isolated_changes": analysis.get("isolated_changes") or [],
+            "validated_changes": analysis.get("validated_changes") or [],
+        }
+        return out
+
+    # Build score map from ranked_nodes (already sorted -score, fqn)
+    score_map: dict[str, float] = {rn["entity"]: rn["score"] for rn in ranked}
+    kept_fqns: Optional[set[str]] = None
+
+    if max_nodes is not None and len(nodes) > max_nodes:
+        nodes_sorted = sorted(
+            nodes,
+            key=lambda n: (-score_map.get(n["fqn"], 0.0), n["fqn"]),
+        )
+        nodes = nodes_sorted[:max_nodes]
+        kept_fqns = {n["fqn"] for n in nodes}
+        ranked = [rn for rn in ranked if rn["entity"] in kept_fqns]
+
+    if kept_fqns is not None or max_edges is not None:
+        if kept_fqns is not None:
+            # Priority: edges where both endpoints are kept nodes
+            priority = [e for e in edges if e["from"] in kept_fqns and e["to"] in kept_fqns]
+            rest = [e for e in edges if not (e["from"] in kept_fqns and e["to"] in kept_fqns)]
+            edges = priority + rest
+        if max_edges is not None:
+            edges = edges[:max_edges]
+
+    out["graph"] = {"nodes": nodes, "edges": edges}
+    out["impact"] = {
+        "global_score": (ir.get("impact") or {}).get("global_score", 0),
+        "ranked_nodes": ranked,
+    }
+    return out
+
+
 # ---------------------------------------------------------------------------
 # Convenience: find Java files in a repo
 # ---------------------------------------------------------------------------
 
 def find_java_files(root: Path, *, max_files: int = 500) -> list[str]:
-    """Return relative paths to Java files under root, excluding test dirs."""
+    """Return relative paths to Java files under root, excluding test dirs and vendor."""
     results: list[str] = []
     for p in sorted(root.rglob("*.java")):
         if len(results) >= max_files:
@@ -1556,7 +1645,12 @@ def find_java_files(root: Path, *, max_files: int = 500) -> list[str]:
             rel = str(p.relative_to(root)).replace("\\", "/")
         except ValueError:
             continue
+        parts = rel.split("/")
+        # Skip test dirs
         if "/test/" in rel or "/tests/" in rel or rel.startswith("test/"):
             continue
+        # Skip vendor/generated/build dirs
+        if any(part in _VENDOR_DIRS for part in parts[:-1]):
+            continue
         results.append(rel)
     return results

{sourcecode-1.30.17.dist-info → sourcecode-1.30.19.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sourcecode
-Version: 1.30.17
+Version: 1.30.19
 Summary: Deterministic codebase context for AI coding agents
 License:                                  Apache License
                                    Version 2.0, January 2004
@@ -221,7 +221,7 @@ Description-Content-Type: text/markdown
 
 **Deterministic, behavior-aware codebase context for AI agents and PR review.**
 
-![Version](https://img.shields.io/badge/version-1.30.17-blue)
+![Version](https://img.shields.io/badge/version-1.30.19-blue)
 ![Python](https://img.shields.io/badge/python-3.10%2B-green)
 
 ---
@@ -257,7 +257,7 @@ pipx install sourcecode
 
 ```bash
 sourcecode version
-# sourcecode 1.30.17
+# sourcecode 1.30.19
 ```
 
 ---

{sourcecode-1.30.17.dist-info → sourcecode-1.30.19.dist-info}/RECORD

@@ -1,16 +1,16 @@
-sourcecode/__init__.py,sha256=C0dqdKsiSL2QvbpNXAjIuRvmS92c_QLcSsSuvX42hRE,104
+sourcecode/__init__.py,sha256=INKPCnhh5ICq3yZ7MYdKW9iFC8jUinfGzMfJk94LSC0,104
 sourcecode/adaptive_scanner.py,sha256=RTNExwWPXzjgLaRueT7UuxkPj5ZEToWjGbx1j0LSZ9E,10250
 sourcecode/architecture_analyzer.py,sha256=MyBa0Hf5HmkudZQDLKrjcWDKETXETXl0mQX1swtTwAA,39091
 sourcecode/architecture_summary.py,sha256=z34_6v7cSwy98cof2UVciGho7SCrZ93tiqMmq5WNzRQ,20405
 sourcecode/ast_extractor.py,sha256=XgrZg2DcWcUm9r87cRG3KGO7IK2TIL_N-CvhSbUmmh4,49901
 sourcecode/classifier.py,sha256=pYve2J1LqtYssU3lYLMDz18PT-CjN5c18QYE7R_IG1Q,7507
-sourcecode/cli.py,sha256=SQfchQoeVZB7ATLOTL_fimtgqmUORZE6rUG-vSiySJc,85829
+sourcecode/cli.py,sha256=an8qIJRnWekBUSyaLod6pD7VyU56V38EotFHDnXlNs8,87996
 sourcecode/code_notes_analyzer.py,sha256=y1MJBnPZHYp4i6cQCXUb9ATIyifS_qMQWjw_8lPkpsU,9215
 sourcecode/confidence_analyzer.py,sha256=xw_Jv8pAd0wd8t2vvQlorw8Ih0rSF3YCoFS8K-_4aXg,15762
 sourcecode/context_scorer.py,sha256=QpChSpsmaAYz91rXA4Ue5xzQmNz_ZboZN09YOHScq1U,14679
 sourcecode/context_summarizer.py,sha256=CiQrfBEzun949bWvmLabWoj2HhPn6Lw62ofqnsy0FlQ,6503
 sourcecode/contract_model.py,sha256=nRxJKPMs1VHwFTa8AVXhGmaLjti3Lr2sjHDpWgv1bfE,3917
-sourcecode/contract_pipeline.py,sha256=VEOvmj-emKuUT-GUosde9FRYyeH8szNW1WSnisNSs2o,27592
+sourcecode/contract_pipeline.py,sha256=w18t_MdbrkIeLcCW-VMQYeb9hlWenAOB-NMiME_Fo-Y,27652
 sourcecode/coverage_parser.py,sha256=q0LeZJaX1bnntLu-ImksdBsMlpsVmk_iUfSaB4eaJGo,19702
 sourcecode/dependency_analyzer.py,sha256=p4ljXhkcGBbFlhaZuPrsjOVjDXaKLTg0Gor2p4qFPP0,56208
 sourcecode/doc_analyzer.py,sha256=afA4uJFwXZ_uR2l4J0pQwbeTkRkGmKdN9KhRVYePBUw,24331
@@ -18,17 +18,17 @@ sourcecode/entrypoint_classifier.py,sha256=gvKgl0f5T8ol1r4JMmkeqGHuZTfZJiOwFOWdc
 sourcecode/env_analyzer.py,sha256=GxCidahAAIptTdDFIlVB6URd4HBnBlIX_SqUov3MBRQ,22076
 sourcecode/file_classifier.py,sha256=48ly5Z6exkzBy8lNy1AkdP4-oJqIA1zT3LZfffuTyDo,11572
 sourcecode/flow_analyzer.py,sha256=dSiuY4w49k29jW_EPXUOND9B5uVbuCA7kjnuHi-pIWA,28781
-sourcecode/git_analyzer.py,sha256=_pCg2V4d2aa17k9hayTzpexAj8syvyk4y9NYNvvgOAI,12802
+sourcecode/git_analyzer.py,sha256=0Gyj-vMpIIN4nfriKXVRouNYBeJ59s6pQDX2Xu9Pq-U,13177
 sourcecode/graph_analyzer.py,sha256=iUK-7pSV-cvGqqD2hENdYmhnm0wcXFEyK-xnu5ul8OU,62515
 sourcecode/metrics_analyzer.py,sha256=m0ENgtqKeBL17kUIK3fmGkgo7UfXBNHxCMj0H_Y5K7c,22750
 sourcecode/pr_comment_renderer.py,sha256=k5pCIP6iBNwy5UYPxu47CAq-62j4E9QZZOPyL3trH80,14799
-sourcecode/prepare_context.py,sha256=pStv1z7bO0_p7YmUTH6NSnc7xjmkNrA9Atdvp6y8UiY,152165
+sourcecode/prepare_context.py,sha256=8-xithuC0-HG12tF6qWW7-3vNtZET9UC2NRbuJkaFnE,158731
 sourcecode/progress.py,sha256=qn30sWaHOkjTgXsSBmiPkz7Rsbwc5oSlIe6JNEMYp_k,3149
 sourcecode/ranking_engine.py,sha256=virVglafZufioHpZpwktjMvUiL0TZELWQCQnQNV8dFo,9360
 sourcecode/redactor.py,sha256=xuGcadGEHaPw4qZXlMDvzMCsr4VOkdp3oBQptHyJk8c,2884
 sourcecode/relevance_scorer.py,sha256=MYF4FFkveAQps9SmTeTlh6ODiBz2F--_hWNeHMLtUHQ,8405
 sourcecode/repo_classifier.py,sha256=FG1vaWKdWXsWdl-S8hjVMiTqcwgaRXkDyvK4rPcOGtQ,22681
-sourcecode/repository_ir.py,sha256=uwi6-mJPh1E2-RYjf-ndqUXTvN32-7Nfv6oFqwHQi1c,57566
+sourcecode/repository_ir.py,sha256=e9TzNbfo_2Cyh3aASFZBq--rPjBMIYMuw5smIWJ5hx8,61099
 sourcecode/runtime_classifier.py,sha256=zWX3r3HCKHc-qtIobErOa8aKMmaoPYREtJKvPcBGPjQ,14792
 sourcecode/scanner.py,sha256=aM3h9-DCQ3xKpeHpHYdo2vX6T5P95HA_YwZbkAVNwmo,8288
 sourcecode/schema.py,sha256=fj3BZ3IcnNV4j21BFIEvz8Qnw_vZoqIbzzRg-qQ-nd0,24530
@@ -64,8 +64,8 @@ sourcecode/telemetry/consent.py,sha256=wLMvGNJeSSyZoNkQXpoUioY6mMv4Qdvuw7S9jAEWn
 sourcecode/telemetry/events.py,sha256=oEvvulfsv5GIDWG2174gSS6tNB95w38AIYiYeifGKlE,2294
 sourcecode/telemetry/filters.py,sha256=Asa71oRl7q3Wt_FMwuufIZJFzSYdgRNKS8LHCIyFeYE,4805
 sourcecode/telemetry/transport.py,sha256=KJeIPCPWMdmbCP3ySGs2iUlia34U6vWne2dZsUezesw,1560
-sourcecode-1.30.17.dist-info/METADATA,sha256=4flsXx7vygTH8S2_G_Dfq1CocZivzyUNQFLBNxY90ZY,28956
-sourcecode-1.30.17.dist-info/WHEEL,sha256=QccIxa26bgl1E6uMy58deGWi-0aeIkkangHcxk2kWfw,87
-sourcecode-1.30.17.dist-info/entry_points.txt,sha256=ex3F9rmbXeyDIoFQHtkEqTsKSaJow8F0LrVu8XfIktQ,57
-sourcecode-1.30.17.dist-info/licenses/LICENSE,sha256=7DdHrU9Z_3e7dSvq4ISijZNjnuHo5NIHNiHDouMQ9JU,10491
-sourcecode-1.30.17.dist-info/RECORD,,
+sourcecode-1.30.19.dist-info/METADATA,sha256=MNkVBrHuR2xR9onBA950wqcFlHu13VRcjAjRQKsqlAo,28956
+sourcecode-1.30.19.dist-info/WHEEL,sha256=QccIxa26bgl1E6uMy58deGWi-0aeIkkangHcxk2kWfw,87
+sourcecode-1.30.19.dist-info/entry_points.txt,sha256=ex3F9rmbXeyDIoFQHtkEqTsKSaJow8F0LrVu8XfIktQ,57
+sourcecode-1.30.19.dist-info/licenses/LICENSE,sha256=7DdHrU9Z_3e7dSvq4ISijZNjnuHo5NIHNiHDouMQ9JU,10491
+sourcecode-1.30.19.dist-info/RECORD,,