contexta-ai 2.2.0__tar.gz → 2.2.1__tar.gz

contexta_ai-2.2.1/.github/workflows/ci.yml

@@ -0,0 +1,73 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ci-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test:
+    name: tests (${{ matrix.os }}, Python ${{ matrix.python }})
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, windows-latest, macos-latest]
+        python: ["3.11", "3.13"]
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0 # the graph and `--for review` read git history
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python }}
+          cache: pip
+          cache-dependency-path: |
+            pyproject.toml
+            requirements-dev.txt
+
+      - name: Linux GUI dependencies
+        if: runner.os == 'Linux'
+        run: sudo apt-get update && sudo apt-get install -y --no-install-recommends xvfb python3-tk
+
+      - name: Install
+        run: python -m pip install --upgrade pip && python -m pip install -e . -r requirements-dev.txt
+
+      - name: Test (Linux, virtual display)
+        if: runner.os == 'Linux'
+        run: xvfb-run -a python -m pytest -q
+
+      - name: Test
+        if: runner.os != 'Linux'
+        run: python -m pytest -q
+
+  wheel:
+    name: wheel installs and runs
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - name: Build the wheel, install it in a clean virtualenv, run the CLI and the MCP server
+        run: python -m pip install --upgrade pip build && python scripts/smoke_wheel.py
+
+  lint:
+    name: ruff (informational)
+    runs-on: ubuntu-latest
+    continue-on-error: true # many pre-existing findings; shown, not enforced yet
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - run: python -m pip install ruff
+      - run: ruff check contexta_app contexta.py contexta_mcp.py scripts tests --output-format=github

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/.github/workflows/publish-pypi.yml

@@ -29,5 +29,8 @@ jobs:
       - name: Build package
         run: python -m build
 
+      - name: Smoke-test the wheel in a clean virtualenv
+        run: python scripts/smoke_wheel.py --wheel "$(ls dist/*.whl)"
+
       - name: Publish to PyPI
         uses: pypa/gh-action-pypi-publish@release/v1

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/CHANGELOG.md

@@ -6,6 +6,28 @@ Versioning: [Semantic Versioning](https://semver.org/)
 
 ---
 
+## [2.2.1] - 2026-10-24
+
+A fix release. **If you installed 2.2.0 or 2.0.0 from PyPI, `contexta` did not start** (`ModuleNotFoundError: No module named 'contexta'`); upgrade.
+
+### Fixed
+- **The PyPI wheel works.** The console script lived in a top-level module the wheel did not include. `main()` is now `contexta_app/entry.py`, `contexta.py` is a thin launcher, and `python -m contexta_app` works. `scripts/smoke_wheel.py` builds the wheel, installs it in a clean virtualenv and checks `contexta --version`, a pack and `contexta serve` over MCP stdio; the publish workflow runs it before uploading
+- `contexta serve --help` printed "unknown option '--help'" and now prints usage
+- The desktop app's editor MCP setup raised `NameError: messagebox` after writing the config. The config it writes now works from the packaged app (`<exe> serve`) and from a pip install (`python -m contexta_app serve`), and an editor config that cannot be parsed is copied to `.bak` instead of being replaced
+- A directory symlink inside the project (`self -> .`) was followed until the 2000-file cap, listing the same files thousands of times. Directory symlinks that stay inside the project are no longer walked; file symlinks and the rule that links leaving the project are ignored are unchanged
+- The project graph shared the 2000-file cap of a pack and hid it, so on a larger repo `graph impact` answered "(none found in graph)". The graph now has its own limit (50000, or `CONTEXTA_GRAPH_MAX_FILES`), and `graph update` and `graph status` warn when it is reached
+- Any desktop Python project was described as "Scan local repositories and generate curated context packs", Contexta's own purpose
+- Python files are now read with `ast` whether or not the optional tree-sitter is installed. Without it, `from .util import slugify` was seen as importing `.util` only, and with it as importing `slugify` as if it were a module (relative imports never resolved to a local file). `--focus slugify` therefore selected different files depending on what happened to be installed. Modules and imported names are now kept apart, and `--focus` matches both
+
+### Added
+- **Token budget** box in the desktop app (`8000`, `8,000` or `32k`), using the same fitter as `--budget`
+- `border_strong` in the theme: inputs, ghost buttons, combo boxes and switch tracks reach 3:1 against their background in both themes
+- CI (`.github/workflows/ci.yml`): the test suite on Linux, Windows and macOS, a wheel smoke test, and ruff in non-blocking mode
+
+### Changed
+- Scan options in the desktop app use one column, so no label is cut off. Entry and Text widgets no longer draw the default white focus ring
+- SECURITY.md and the README now say what touches the network: the one-time download of the `cl100k_base` token encoding, the optional `--ask`, and the GitHub Action's PR comment
+
 ## [2.2.0] - 2026-10-24
 
 Contexta now remembers your project. A local graph of files, symbols, tests and git history answers

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/CONTRIBUTING.md

@@ -8,6 +8,7 @@ Thank you for considering a contribution!
 - Optional packaging tools such as **Nuitka** or **PyInstaller** must stay out of runtime code paths
 - Python **3.10+** syntax is acceptable
 - Every PR must keep the project runnable with `python contexta.py` after `python -m pip install -r requirements.txt`
+- Tests: `python -m pip install -e . -r requirements-dev.txt`, then `python -m pytest -q` (on Linux add `xvfb-run -a` for the GUI tests; the tests that need Pillow or PyYAML skip when those are missing)
 
 ## How to contribute
 

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: contexta-ai
-Version: 2.2.0
+Version: 2.2.1
 Summary: AI-native codebase context engine. Scans projects and serves intelligent context packs to Claude Code, Cursor, Windsurf, and any MCP-compatible tool.
 Project-URL: Homepage, https://github.com/pablokaua03/contexta
 Project-URL: Repository, https://github.com/pablokaua03/contexta
@@ -58,7 +58,7 @@ Contexta reads a codebase, keeps the files that matter for the task, fits them i
 [![Python](https://img.shields.io/badge/python-3.11%2B-12211c?style=flat-square)](https://python.org)
 [![License: MIT](https://img.shields.io/badge/license-MIT-12211c?style=flat-square)](LICENSE)
 [![Platform](https://img.shields.io/badge/platform-Windows%20%7C%20macOS%20%7C%20Linux-12211c?style=flat-square)]()
-[![Version](https://img.shields.io/badge/version-2.2.0-12211c?style=flat-square)]()
+[![Version](https://img.shields.io/badge/version-2.2.1-12211c?style=flat-square)]()
 [![MCP](https://img.shields.io/badge/MCP-compatible-12211c?style=flat-square)]()
 
 <br>
@@ -429,7 +429,7 @@ python -m pytest tests/ -k "not test_relation_score"
 ## Security and behavior
 
 - Read-only, with one exception: Contexta never modifies your files. It does keep its project graph in `.contexta/` (names, signatures and relations, never file contents; it ignores itself with a `*` `.gitignore`). Turn it off with `--no-graph` or `graph = false`, move it with `CONTEXTA_GRAPH_DIR`
-- No runtime telemetry or network requirement (except optional `--ask` AI calls)
+- No telemetry. Packs, the graph and the MCP server work offline. The only network use: the first token count downloads the `cl100k_base` encoding once (cached by `tiktoken`; offline, Contexta falls back to a character-based estimate), the optional `--ask` calls go to the AI provider you configured, and the GitHub Action posts its PR comment to the GitHub API
 - Scan limit of 2000 files by default, raised with `--max-files`
 - Embedded binary/blob payloads are suppressed automatically
 - Likely secrets are redacted in exports and in MCP `read_files` unless `--no-redact` is set. `--dry-run` shows what would be redacted

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/README.md

@@ -11,7 +11,7 @@ Contexta reads a codebase, keeps the files that matter for the task, fits them i
 [![Python](https://img.shields.io/badge/python-3.11%2B-12211c?style=flat-square)](https://python.org)
 [![License: MIT](https://img.shields.io/badge/license-MIT-12211c?style=flat-square)](LICENSE)
 [![Platform](https://img.shields.io/badge/platform-Windows%20%7C%20macOS%20%7C%20Linux-12211c?style=flat-square)]()
-[![Version](https://img.shields.io/badge/version-2.2.0-12211c?style=flat-square)]()
+[![Version](https://img.shields.io/badge/version-2.2.1-12211c?style=flat-square)]()
 [![MCP](https://img.shields.io/badge/MCP-compatible-12211c?style=flat-square)]()
 
 <br>
@@ -382,7 +382,7 @@ python -m pytest tests/ -k "not test_relation_score"
 ## Security and behavior
 
 - Read-only, with one exception: Contexta never modifies your files. It does keep its project graph in `.contexta/` (names, signatures and relations, never file contents; it ignores itself with a `*` `.gitignore`). Turn it off with `--no-graph` or `graph = false`, move it with `CONTEXTA_GRAPH_DIR`
-- No runtime telemetry or network requirement (except optional `--ask` AI calls)
+- No telemetry. Packs, the graph and the MCP server work offline. The only network use: the first token count downloads the `cl100k_base` encoding once (cached by `tiktoken`; offline, Contexta falls back to a character-based estimate), the optional `--ask` calls go to the AI provider you configured, and the GitHub Action posts its PR comment to the GitHub API
 - Scan limit of 2000 files by default, raised with `--max-files`
 - Embedded binary/blob payloads are suppressed automatically
 - Likely secrets are redacted in exports and in MCP `read_files` unless `--no-redact` is set. `--dry-run` shows what would be redacted

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/README.pt-BR.md

@@ -11,7 +11,7 @@ O Contexta lê o código, fica com os arquivos que importam para a tarefa, encai
 [![Python](https://img.shields.io/badge/python-3.11%2B-12211c?style=flat-square)](https://python.org)
 [![License: MIT](https://img.shields.io/badge/license-MIT-12211c?style=flat-square)](LICENSE)
 [![Plataforma](https://img.shields.io/badge/plataforma-Windows%20%7C%20macOS%20%7C%20Linux-12211c?style=flat-square)]()
-[![Versão](https://img.shields.io/badge/vers%C3%A3o-2.2.0-12211c?style=flat-square)]()
+[![Versão](https://img.shields.io/badge/vers%C3%A3o-2.2.1-12211c?style=flat-square)]()
 [![MCP](https://img.shields.io/badge/MCP-compat%C3%ADvel-12211c?style=flat-square)]()
 
 <br>
@@ -366,7 +366,7 @@ python -m pytest tests/ -k "not test_relation_score"
 ## Segurança e comportamento
 
 - Somente leitura, com uma exceção: o Contexta nunca modifica seus arquivos. Ele guarda o grafo do projeto em `.contexta/` (nomes, assinaturas e relações, nunca o conteúdo dos arquivos; ele se ignora com um `.gitignore` com `*`). Desligue com `--no-graph` ou `graph = false`, mude o local com `CONTEXTA_GRAPH_DIR`
-- Sem telemetria ou necessidade de rede (exceto `--ask` opcional)
+- Sem telemetria. Packs, grafo e servidor MCP funcionam offline. Único uso de rede: a primeira contagem de tokens baixa a codificação `cl100k_base` uma vez (guardada em cache pelo `tiktoken`; sem rede, o Contexta usa uma estimativa por caracteres), as chamadas opcionais de `--ask` vão para o provedor de IA que você configurou, e a GitHub Action publica o comentário do PR na API do GitHub
 - Limite de 2000 arquivos por padrão, ajustável com `--max-files`
 - Payloads binários/base64 suprimidos automaticamente
 - Segredos prováveis são redatados nos exports e no `read_files` do MCP, a menos que `--no-redact` seja usado. `--dry-run` mostra o que seria redatado

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/SECURITY.md

@@ -21,10 +21,11 @@ You can expect a response within 7 days.
 
 ## Security notes
 
-- `Contexta` is a read-only tool; it does not modify, delete, or upload project files
-- It makes no network requests of any kind
+- `Contexta` never modifies, deletes, or uploads your project files. It does write its own project graph to `.contexta/` (or `CONTEXTA_GRAPH_DIR`) unless `--no-graph` is set, and `contexta install` writes agent rules and MCP config files you ask for
+- It sends no telemetry. Packs, the project graph and the MCP server do not need the network
+- The network is used in three cases only: the first token count downloads the `cl100k_base` encoding once through `tiktoken` (cached afterwards; offline, Contexta falls back to a character-based estimate), the optional `--ask` feature sends the pack to the AI provider you configured, and the GitHub Action posts its pull request comment to the GitHub API
 - It uses local runtime dependencies for parsing, token estimation, and repository analysis
 - On some Linux distributions, `tkinter` may come from a system package such as `python3-tk`
 - Windows builds use `Nuitka`; Linux packaging uses `PyInstaller`
-- All processing happens locally on your machine
+- Analysis happens locally on your machine
 - Exports replace likely secrets (API keys, private keys, tokens, connection-string passwords) with `[REDACTED]` before the pack is written. This is best-effort and does not make it safe to commit real credentials

contexta_ai-2.2.1/contexta.py

@@ -0,0 +1,20 @@
+"""
+Contexta — AI-native codebase context engine.
+
+Usage:
+    contexta serve [--root DIR]  Start the MCP server (for Claude Code, Cursor, etc.).
+                                 Tools only read inside --root (repeatable), CONTEXTA_ROOT,
+                                 or the current directory
+    contexta graph <command>    Ask the project graph: dependents, impact, hot files, repo map
+    contexta install <target>   Write agent rules or MCP config (cursor, claude, copilot, mcp)
+    contexta <project> [opts]   Generate a context pack (CLI mode)
+    contexta                    Launch the desktop GUI
+    contexta --configure-ai     Set up AI API keys
+"""
+
+from __future__ import annotations
+
+from contexta_app.entry import main
+
+if __name__ == "__main__":
+    main()

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/contexta_app/__init__.py

@@ -2,4 +2,4 @@
 Contexta application package.
 """
 
-__version__ = "2.2.0"
+__version__ = "2.2.1"

contexta_ai-2.2.1/contexta_app/__main__.py

@@ -0,0 +1,4 @@
+from contexta_app.entry import main
+
+if __name__ == "__main__":
+    main()

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/contexta_app/context_engine.py

@@ -124,10 +124,12 @@ from contexta_app.engine.relations import (  # noqa: F401
     test_relation_score,
 )
 from contexta_app.engine.insight import (  # noqa: F401
+    _python_ast_symbols,
     collect_file_signals,
     extract_symbols,
     flatten_tree,
     make_file_insight,
+    python_imported_names,
 )
 from contexta_app.engine.roles import (  # noqa: F401
     _CONTEXTA_MODULE_ROLES,

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/contexta_app/engine/fingerprint.py

@@ -326,11 +326,9 @@ def infer_probable_purpose(project_type_label: str | None, domain_label: str | N
         ("Electron desktop application", "image_editing"): "Provide desktop image or asset-editing workflows through a web-powered desktop shell.",
         ("Go application", "task_management"): "Provide Go-based services or tooling for task, board, and workflow management operations.",
         ("Rust application", "graph_analysis"): "Analyze, transform, or serve graph and network data with performance-oriented Rust modules.",
-        ("Desktop GUI + CLI developer tool", "developer_tooling"): "Scan local repositories and generate curated context packs for AI workflows.",
+        ("Desktop GUI + CLI developer tool", "developer_tooling"): "Provide developer tooling, automation, or repository-analysis workflows through a desktop GUI and a command-line entry point.",
         ("Desktop GUI + CLI developer tool", "image_editing"): "Provide desktop editing workflows for image or asset manipulation through GUI and command-driven tooling.",
     }
-    if project_type_label == "Desktop GUI + CLI developer tool":
-        return "Scan local repositories and generate curated context packs for AI workflows."
     if (project_type_label, domain_label) in purpose_by_pair:
         return purpose_by_pair[(project_type_label, domain_label)]
 
@@ -418,6 +416,8 @@ def infer_probable_purpose(project_type_label: str | None, domain_label: str | N
         return "Build JVM-based application logic using Scala and functional or object-oriented patterns."
     if project_type_label == "PHP application":
         return "Deliver web application functionality using PHP modules and associated server-side tooling."
+    if project_type_label == "Desktop GUI + CLI developer tool":
+        return "Provide a desktop application with a graphical interface and a command-line entry point."
     if project_type_label == "Python application":
         return "Organize Python modules, scripts, and supporting logic into a runnable application."
     if project_type_label == "JavaScript/TypeScript application":

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/contexta_app/engine/focus.py

@@ -32,7 +32,7 @@ def compute_focus_raw(item: FileInsight, query: str) -> float:
     score = 0.0
     rel = item.relpath.as_posix().lower()
     stem = item.path.stem.lower()
-    imports_blob = " ".join(imported.lower() for imported in item.imports)
+    imports_blob = " ".join(imported.lower() for imported in (*item.imports, *item.imported_names))
     functions_blob = " ".join(func.lower() for func in item.functions)
     classes_blob = " ".join(cls.lower() for cls in item.classes)
     content_lower = item.content_lower

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/contexta_app/engine/insight.py

@@ -2,7 +2,9 @@
 
 from __future__ import annotations
 
+import ast
 import re
+from functools import lru_cache
 from pathlib import Path
 from contexta_app.scanner import get_language, read_file_safe
 from contexta_app.syntax import extract_symbols_with_treesitter
@@ -22,6 +24,7 @@ def make_file_insight(project_path: Path, path: Path) -> FileInsight:
     content, truncated, total_line_count = read_file_safe(path)
     lang = get_language(path)
     functions, classes, imports = extract_symbols(content, lang)
+    imported_names = python_imported_names(content) if lang == "python" else []
     external_imports = [item for item in imports if not item.startswith(".")]
     return FileInsight(
         path=path,
@@ -34,10 +37,46 @@ def make_file_insight(project_path: Path, path: Path) -> FileInsight:
         functions=functions,
         classes=classes,
         imports=imports,
+        imported_names=imported_names,
         external_imports=external_imports,
     )
 
+@lru_cache(maxsize=64)
+def _python_ast_symbols(content: str) -> tuple[tuple[str, ...], tuple[str, ...], tuple[str, ...], tuple[str, ...]] | None:
+    """Top-level functions and classes, imported modules and imported names, from ``ast``.
+
+    Python is read with the standard library whether or not tree-sitter is installed, so a pack
+    does not depend on an optional package. Modules are written the way ``resolve_local_import``
+    expects them: ``pkg.core``, or leading dots for a relative import (``.util``).
+    """
+    try:
+        tree = ast.parse(content)
+    except (SyntaxError, ValueError, RecursionError):
+        return None
+    functions = [n.name for n in tree.body if isinstance(n, (ast.FunctionDef, ast.AsyncFunctionDef))]
+    classes = [n.name for n in tree.body if isinstance(n, ast.ClassDef)]
+    modules: list[str] = []
+    names: list[str] = []
+    for node in ast.walk(tree):
+        if isinstance(node, ast.Import):
+            modules.extend(alias.name for alias in node.names)
+        elif isinstance(node, ast.ImportFrom):
+            modules.append("." * (node.level or 0) + (node.module or ""))
+            names.extend(alias.name for alias in node.names if alias.name != "*")
+    unique = lambda values, limit: tuple(dict.fromkeys(values))[:limit]  # noqa: E731
+    return unique(functions, 12), unique(classes, 8), unique(modules, 24), unique(names, 40)
+
+def python_imported_names(content: str) -> list[str]:
+    """Names brought in with ``from x import name``, for focus matching ("slugify" finds its callers)."""
+    parsed = _python_ast_symbols(content)
+    return list(parsed[3]) if parsed else []
+
 def extract_symbols(content: str, lang: str | None) -> tuple[list[str], list[str], list[str]]:
+    if lang == "python":
+        parsed = _python_ast_symbols(content)
+        if parsed:
+            return list(parsed[0]), list(parsed[1]), list(parsed[2])
+
     syntax_symbols = extract_symbols_with_treesitter(content, lang)
     if syntax_symbols:
         functions, classes, imports = syntax_symbols

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/contexta_app/engine/models.py

@@ -36,6 +36,7 @@ class FileInsight:
     functions: list[str] = field(default_factory=list)
     classes: list[str] = field(default_factory=list)
     imports: list[str] = field(default_factory=list)
+    imported_names: list[str] = field(default_factory=list)  # Python: names from `from x import name`
     local_imports: list[str] = field(default_factory=list)
     external_imports: list[str] = field(default_factory=list)
     tags: set[str] = field(default_factory=set)

contexta_ai-2.2.0/contexta.py → contexta_ai-2.2.1/contexta_app/entry.py

@@ -1,4 +1,6 @@
-"""
+"""Command-line entry point shared by the `contexta` console script, `python -m contexta_app` and the
+legacy `contexta.py` launcher.
+
 Contexta — AI-native codebase context engine.
 
 Usage:

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/contexta_app/graph.py

@@ -99,6 +99,7 @@ class UpdateStats:
     seconds: float = 0.0
     git_commits: int = 0
     skipped_reason: str = ""
+    truncated: bool = False
 
     def to_dict(self) -> dict:
         return {k: (round(v, 3) if isinstance(v, float) else v) for k, v in self.__dict__.items()}
@@ -106,15 +107,31 @@ class UpdateStats:
     def summary(self) -> str:
         if self.skipped_reason:
             return f"Graph not updated: {self.skipped_reason}"
-        return (
+        text = (
             f"Graph: {self.total} file(s), {self.added} new, {self.changed} changed, {self.removed} removed, "
             f"{self.unchanged} unchanged, {self.symbols} symbols, {self.edges} edges ({self.seconds:.2f}s)"
         )
+        if self.truncated:
+            text += f"\nWarning: the project has more than {self.total} files; only the first {self.total} were indexed. Raise CONTEXTA_GRAPH_MAX_FILES to index the rest."
+        return text
 
 
 # ------------------------------------------------------------------ helpers
 
 
+DEFAULT_GRAPH_MAX_FILES = 50000
+
+
+def graph_max_files() -> int:
+    """File cap for the project graph: CONTEXTA_GRAPH_MAX_FILES, else 50000 (packs keep their own 2000 default)."""
+    raw = os.environ.get("CONTEXTA_GRAPH_MAX_FILES", "").strip()
+    try:
+        value = int(raw) if raw else 0
+    except ValueError:
+        value = 0
+    return value if value > 0 else DEFAULT_GRAPH_MAX_FILES
+
+
 def classify_kind(rel: str, lang: str | None) -> str:
     parts = PurePosixPath(rel).parts
     name = parts[-1]
@@ -350,6 +367,7 @@ class ProjectGraph:
         self._shared = explicit_db and self.db_path.parent != self.root / GRAPH_DIRNAME
         self.readonly = readonly
         self._con: sqlite3.Connection | None = None
+        self.truncated = False
 
     # connection
     @property
@@ -427,6 +445,7 @@ class ProjectGraph:
         started = time.perf_counter()
         stats = UpdateStats()
         self._claim()
+        self.truncated = False
         if files is None:
             files = self._scan()
         current: dict[str, Path] = {}
@@ -439,6 +458,7 @@ class ProjectGraph:
                 continue
             current[rel] = path
         stats.total = len(current)
+        stats.truncated = self.truncated
 
         con = self.con
         known = {r["path"]: r for r in con.execute("SELECT path, hash, mtime_ns, size FROM files")}
@@ -485,6 +505,7 @@ class ProjectGraph:
             if dirty or use_git:
                 self._compute_ranks()
             self._set_meta("updated_at", str(now))
+            self._set_meta("truncated", "1" if self.truncated else "0")
             con.commit()
         except Exception:
             con.rollback()
@@ -532,7 +553,11 @@ class ProjectGraph:
         from contexta_app.scanner import build_tree, load_gitignore_patterns
 
         patterns = load_gitignore_patterns(self.root)
-        tree = build_tree(self.root, False, False, lambda *_a, **_k: None, [0], patterns, self.root)
+        # The memory should cover the whole project, so it does not share the 2000-file cap of a pack.
+        limit = graph_max_files()
+        notes: list[str] = []
+        tree = build_tree(self.root, False, False, lambda msg, *_a, **_k: notes.append(str(msg)), [0], patterns, self.root, limit)
+        self.truncated = any("file limit" in n for n in notes)
         return flatten_tree(tree)
 
     def _delete_file(self, rel: str) -> None:
@@ -785,6 +810,7 @@ class ProjectGraph:
             "updated_at": _iso(float(self.meta("updated_at", "0") or 0)),
             "db": str(self.db_path.relative_to(self.root)) if self.db_path.is_relative_to(self.root) else "(outside project)",
             "parsers": {r[0] or "none": r[1] for r in con.execute("SELECT backend, COUNT(*) FROM files GROUP BY backend")},
+            "truncated": self.meta("truncated", "0") == "1",
         }
 
     def has_file(self, path: str) -> bool:

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/contexta_app/graph_cli.py

@@ -103,6 +103,8 @@ def format_status(info: dict, name: str) -> str:
     lines.append("  edges: " + (", ".join(f"{k} {v}" for k, v in sorted(info["edges"].items())) or "none"))
     lines.append("  parsers: " + ", ".join(f"{k} {v}" for k, v in sorted(info["parsers"].items())))
     lines.append(f"  updated {info['updated_at']}   db {info['db']}")
+    if info.get("truncated"):
+        lines.append("  warning: the last update hit the file limit, so the graph is incomplete (raise CONTEXTA_GRAPH_MAX_FILES)")
     return "\n".join(lines) + "\n"
 
 

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/contexta_app/mcp_server.py

@@ -55,6 +55,9 @@ def parse_serve_args(args: list[str]) -> list[str]:
     i = 0
     while i < len(args):
         arg = args[i]
+        if arg in ("-h", "--help"):
+            print("Usage: contexta serve [--root DIR]\n\nStart the MCP server over stdio. Tools only read inside --root\n(repeatable), CONTEXTA_ROOT, or the current directory.")
+            raise SystemExit(0)
         if arg == "--root":
             if i + 1 >= len(args):
                 raise SystemExit("contexta serve: --root needs a folder")

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/contexta_app/scanner.py

@@ -561,6 +561,11 @@ def build_tree(
 
         if entry.is_symlink() and not _symlink_stays_inside(entry, project_root):
             continue
+        if entry.is_symlink() and entry.is_dir():
+            # A directory link that stays inside the project points at something scanned under its real
+            # path anyway. Following it only lists the same files twice (or forever for `self -> .`, until
+            # the file cap), and could walk a folder the ignore rules skip.
+            continue
 
         if entry.is_dir():
             if should_ignore_dir(entry.name, include_hidden):

{contexta_ai-2.2.0 → contexta_ai-2.2.1}/contexta_app/theme.py

@@ -4,6 +4,10 @@ theme.py — Color palettes, theme switching, and widget registry.
 The palette is an ink surface with a single mint accent. Text colors are
 chosen so body copy and captions stay above a 4.5:1 contrast ratio on the
 surfaces they sit on. Bright fills use dark ink for their label.
+
+``border`` is the quiet outline of a card. ``border_strong`` is for anything the
+user interacts with (inputs, ghost buttons, switch tracks): at least 3:1 against
+the card, as WCAG 1.4.11 asks for UI components.
 """
 
 import tkinter as tk
@@ -20,6 +24,7 @@ DARK: dict[str, str] = {
     "input":       "#0f1613",
     "border":      "#2c433b",
     "border2":     "#3e5c50",
+    "border_strong": "#4f7a6a",
     "accent":      "#3dcc8e",
     "accent_dk":   "#1f8f62",
     "green":       "#3dcc8e",
@@ -49,6 +54,7 @@ LIGHT: dict[str, str] = {
     "input":       "#ffffff",
     "border":      "#d5e2db",
     "border2":     "#b7c9c0",
+    "border_strong": "#728a7d",
     "accent":      "#0d7a4e",
     "accent_dk":   "#0b6842",
     "green":       "#0d7a4e",