aria-code 0.112.0__tar.gz → 0.113.0__tar.gz

{aria_code-0.112.0/src/aria_code.egg-info → aria_code-0.113.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aria-code
-Version: 0.112.0
+Version: 0.113.0
 Summary: Open-source AI agent for coding, financial analysis and logistics operations: numbers you can check, client data kept apart, actions a person approves
 License: Apache-2.0
 Project-URL: Homepage, https://github.com/artheras/aria-code
@@ -171,7 +171,7 @@ Numbers you can check, client data kept apart, actions a person approves.</p>
 <p align="center">
   <a href="https://www.npmjs.com/package/@artheras/aria-code"><img src="https://img.shields.io/npm/v/@artheras/aria-code/latest?style=flat-square&logo=npm&label=npm" alt="npm version"></a>
   <!-- Release automation pins PyPI: the default endpoint still selects legacy 4.x. -->
-  <a href="https://pypi.org/project/aria-code/0.112.0/"><img src="https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fpypi.org%2Fpypi%2Faria-code%2F0.112.0%2Fjson&amp;query=%24.info.version&amp;prefix=v&amp;style=flat-square&amp;logo=pypi&amp;label=PyPI&amp;color=blue&amp;cacheSeconds=300" alt="PyPI version"></a>
+  <a href="https://pypi.org/project/aria-code/0.113.0/"><img src="https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fpypi.org%2Fpypi%2Faria-code%2F0.113.0%2Fjson&amp;query=%24.info.version&amp;prefix=v&amp;style=flat-square&amp;logo=pypi&amp;label=PyPI&amp;color=blue&amp;cacheSeconds=300" alt="PyPI version"></a>
   <a href="https://github.com/artheras/aria-code/actions/workflows/ci.yml"><img src="https://img.shields.io/github/actions/workflow/status/artheras/aria-code/ci.yml?branch=main&style=flat-square&logo=githubactions&label=CI" alt="CI status"></a>
   <a href="LICENSE"><img src="https://img.shields.io/badge/license-Apache%202.0-64748b?style=flat-square" alt="Apache License 2.0"></a>
 </p>

{aria_code-0.112.0 → aria_code-0.113.0}/README.md

@@ -15,7 +15,7 @@ Numbers you can check, client data kept apart, actions a person approves.</p>
 <p align="center">
   <a href="https://www.npmjs.com/package/@artheras/aria-code"><img src="https://img.shields.io/npm/v/@artheras/aria-code/latest?style=flat-square&logo=npm&label=npm" alt="npm version"></a>
   <!-- Release automation pins PyPI: the default endpoint still selects legacy 4.x. -->
-  <a href="https://pypi.org/project/aria-code/0.112.0/"><img src="https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fpypi.org%2Fpypi%2Faria-code%2F0.112.0%2Fjson&amp;query=%24.info.version&amp;prefix=v&amp;style=flat-square&amp;logo=pypi&amp;label=PyPI&amp;color=blue&amp;cacheSeconds=300" alt="PyPI version"></a>
+  <a href="https://pypi.org/project/aria-code/0.113.0/"><img src="https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fpypi.org%2Fpypi%2Faria-code%2F0.113.0%2Fjson&amp;query=%24.info.version&amp;prefix=v&amp;style=flat-square&amp;logo=pypi&amp;label=PyPI&amp;color=blue&amp;cacheSeconds=300" alt="PyPI version"></a>
   <a href="https://github.com/artheras/aria-code/actions/workflows/ci.yml"><img src="https://img.shields.io/github/actions/workflow/status/artheras/aria-code/ci.yml?branch=main&style=flat-square&logo=githubactions&label=CI" alt="CI status"></a>
   <a href="LICENSE"><img src="https://img.shields.io/badge/license-Apache%202.0-64748b?style=flat-square" alt="Apache License 2.0"></a>
 </p>

{aria_code-0.112.0 → aria_code-0.113.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "aria-code"
-version = "0.112.0"
+version = "0.113.0"
 description = "Open-source AI agent for coding, financial analysis and logistics operations: numbers you can check, client data kept apart, actions a person approves"
 readme = "README.md"
 license = { text = "Apache-2.0" }

{aria_code-0.112.0 → aria_code-0.113.0}/src/aria_code/_version.py

@@ -17,4 +17,4 @@ from __future__ import annotations
 
 __all__ = ["__version__"]
 
-__version__ = "0.112.0"
+__version__ = "0.113.0"

{aria_code-0.112.0 → aria_code-0.113.0}/src/aria_code/apps/cli/tools/write_tools.py

@@ -118,6 +118,19 @@ def _print_lsp_diags(diags: list, console, has_rich: bool, limit: int = 6) -> No
         console.print(f"  [dim]… +{extra} more[/dim]") if has_rich else print(f"  ... +{extra} more")
 
 
+def _client_isolation_error(p: pathlib.Path, content: str) -> str | None:
+    """Why a client's document must not be written as it stands (aria_code.safety.client_isolation).
+
+    A failure inside the check never blocks a write: it is a guard on top of
+    the write path, not a new way for writing to break.
+    """
+    try:
+        from aria_code.safety.client_isolation import check_write
+        return check_write(p, content, [pathlib.Path.cwd()])
+    except Exception:
+        return None
+
+
 def _is_safe(p: pathlib.Path, params: dict | None = None) -> bool:
     if params and params.get("_workspace"):
         from aria_code.workspace.files import WorkspaceSecurity
@@ -463,6 +476,9 @@ def tool_write_file(params: dict) -> dict:
         p = raw_path.resolve()
         if not _is_safe(p, params):
             return {"success": False, "error": f"Access denied: path '{p}' is outside allowed directories"}
+        leak = _client_isolation_error(p, content)
+        if leak:
+            return {"success": False, "error": leak, "data": {"client_isolation": True}}
 
         existed = p.exists()
         desktop = pathlib.Path.home() / "Desktop"
@@ -719,6 +735,9 @@ def tool_edit_file(params: dict) -> dict:
 
         replacements = occurrences if replace_all else 1
         new_content = content.replace(old_str, new_str, -1 if replace_all else 1)
+        leak = _client_isolation_error(p, new_content)
+        if leak:
+            return {"success": False, "error": leak, "data": {"client_isolation": True}}
         store = _change_store()
         change = store.stage(p, new_content, source="edit_file")
         added = len(new_str.splitlines())
@@ -855,6 +874,9 @@ def tool_multi_edit(params: dict) -> dict:
 
         # ── Phase 2: stage + apply once ───────────────────────────────────────
         store = _change_store()
+        leak = _client_isolation_error(p, working)
+        if leak:
+            return {"success": False, "error": leak, "data": {"client_isolation": True}}
         change = store.stage(p, working, source="multi_edit")
         console, has_rich = _ui()
 

aria_code-0.113.0/src/aria_code/safety/client_isolation.py

@@ -0,0 +1,178 @@
+"""Keep one client's document free of other clients' data.
+
+A 3PL's stock file, a fund's holdings export or a merchant ledger often holds
+several clients side by side. A document written *for one of them* (named for
+it: ``report_BETA.md``, ``statement-acme.csv``) must not carry another
+client's name, id or item codes. The prompt says so, and in an eval run the
+agent still copied two notes from Beta's own rows that named Acme and Gamma
+SKUs into Beta's report.
+
+This is the check behind that rule, the way a liquid handler's checker
+tracks what each tip has touched rather than trusting the protocol's
+comments: before a client document is written, its text is scanned for the
+identifiers of every *other* client found in the workspace's data files, and
+the write is refused with a list of what leaked.
+
+Clients are found in CSV files near the document that have an owner column
+(``owner_id``, ``shipper``, ``client``, ``customer``…). A document is a
+client's when its file name names exactly one of them. Code is not checked:
+a script that processes every client legitimately names them all, and so
+does an internal report whose name does not single one out.
+"""
+
+from __future__ import annotations
+
+import csv
+import re
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Iterable
+
+OWNER_COLUMNS = ("owner_id", "owner", "shipper_id", "shipper", "client_id", "client",
+                 "customer_id", "customer", "account_id", "货主", "客户")
+NAME_COLUMNS = ("owner_name", "shipper_name", "client_name", "customer_name", "account_name",
+                "货主名称", "客户名称")
+ITEM_COLUMNS = ("sku", "item", "item_id", "item_code", "product_id", "product_code", "sku_id")
+DOCUMENT_SUFFIXES = {".md", ".markdown", ".txt", ".csv", ".tsv", ".json", ".html", ".htm",
+                     ".xml", ".yaml", ".yml", ".rst", ".eml"}
+MIN_TERM = 3                       # "A" or "AB" as an owner id would match everywhere
+MAX_FILE_BYTES = 5 * 1024 * 1024
+MAX_FILES = 25
+
+
+@dataclass
+class Client:
+    id: str
+    names: set[str] = field(default_factory=set)
+    items: set[str] = field(default_factory=set)
+
+    def terms(self) -> list[str]:
+        return sorted({t for t in {self.id, *self.names, *self.items} if len(t) >= MIN_TERM}, key=len, reverse=True)
+
+
+_CACHE: dict[tuple[str, float, int], dict[str, Client]] = {}
+
+
+def _column(header: Iterable[str], wanted: tuple[str, ...]) -> str | None:
+    lowered = {h.strip().lower(): h for h in header if h}
+    for name in wanted:
+        if name in lowered:
+            return lowered[name]
+    return None
+
+
+def _clients_in(path: Path) -> dict[str, Client]:
+    try:
+        stat = path.stat()
+    except OSError:
+        return {}
+    if stat.st_size > MAX_FILE_BYTES:
+        return {}
+    key = (str(path), stat.st_mtime, stat.st_size)
+    if key in _CACHE:
+        return _CACHE[key]
+    found: dict[str, Client] = {}
+    try:
+        with path.open(newline="", encoding="utf-8-sig", errors="replace") as fh:
+            reader = csv.DictReader(fh)
+            header = reader.fieldnames or []
+            owner = _column(header, OWNER_COLUMNS)
+            if owner:
+                name = _column(header, NAME_COLUMNS)
+                item = _column(header, ITEM_COLUMNS)
+                for row in reader:
+                    cid = (row.get(owner) or "").strip()
+                    if not cid:
+                        continue
+                    client = found.setdefault(cid, Client(cid))
+                    if name and (row.get(name) or "").strip():
+                        client.names.add(row[name].strip())
+                    if item and (row.get(item) or "").strip():
+                        client.items.add(row[item].strip())
+    except (OSError, csv.Error, UnicodeError):
+        found = {}
+    _CACHE[key] = found
+    return found
+
+
+def client_index(roots: Iterable[Path]) -> dict[str, Client]:
+    """Every client named in the CSV data files in *roots* and their immediate subfolders."""
+    index: dict[str, Client] = {}
+    seen = 0
+    for root in dict.fromkeys(Path(r) for r in roots):
+        if not root.is_dir():
+            continue
+        candidates = sorted(root.glob("*.csv")) + sorted(root.glob("*/*.csv"))
+        for path in candidates:
+            if seen >= MAX_FILES:
+                return index
+            seen += 1
+            for cid, client in _clients_in(path).items():
+                merged = index.setdefault(cid, Client(cid))
+                merged.names |= client.names
+                merged.items |= client.items
+    return index if len(index) > 1 else {}
+
+
+def _words(text: str) -> set[str]:
+    return {w.lower() for w in re.split(r"[^0-9A-Za-z一-鿿]+", text) if w}
+
+
+def target_client(path: Path, index: dict[str, Client]) -> str | None:
+    """The one client a document's file name singles out, if it singles out exactly one."""
+    words = _words(Path(path).stem)
+    compact = re.sub(r"[^0-9a-z一-鿿]", "", Path(path).stem.lower())
+    hits = set()
+    for cid, client in index.items():
+        if cid.lower() in words:
+            hits.add(cid)
+            continue
+        for name in client.names:
+            squashed = re.sub(r"[^0-9a-z一-鿿]", "", name.lower())
+            if len(squashed) >= MIN_TERM and squashed in compact:
+                hits.add(cid)
+    return next(iter(hits)) if len(hits) == 1 else None
+
+
+def _mentions(text: str, term: str) -> bool:
+    if re.search(r"[一-鿿]", term):
+        return term in text
+    return re.search(rf"(?<![0-9A-Za-z]){re.escape(term)}(?![0-9A-Za-z])", text, re.IGNORECASE) is not None
+
+
+def foreign_mentions(text: str, target: str, index: dict[str, Client]) -> dict[str, list[str]]:
+    """Other clients' identifiers that appear in *text*, by client."""
+    own = {t.lower() for t in index[target].terms()} if target in index else set()
+    leaks: dict[str, list[str]] = {}
+    for cid, client in sorted(index.items()):
+        if cid == target:
+            continue
+        hits = [t for t in client.terms() if t.lower() not in own and _mentions(text, t)]
+        if hits:
+            leaks[cid] = hits[:4]
+    return leaks
+
+
+def check_write(path: Path | str, content: str, roots: Iterable[Path | str] = ()) -> str | None:
+    """Why this write must be refused, or None when it may go ahead."""
+    path = Path(path)
+    if path.suffix.lower() not in DOCUMENT_SUFFIXES:
+        return None
+    search = [path.parent, *(Path(r) for r in roots)]
+    index = client_index(search)
+    if not index:
+        return None
+    target = target_client(path, index)
+    if target is None:
+        return None
+    leaks = foreign_mentions(content, target, index)
+    if not leaks:
+        return None
+    listed = "; ".join(f"{cid} ({', '.join(terms)})" for cid, terms in leaks.items())
+    return (f"Not written: {path.name} is for client {target}, but it contains other clients' data: {listed}. "
+            f"A client's document holds only that client's data. Remove these, including any that came "
+            f"from notes or comments in {target}'s own rows, and write it again. If this is meant to be an "
+            f"internal cross-client document, name it without singling out one client.")
+
+
+__all__ = ["Client", "check_write", "client_index", "foreign_mentions", "target_client"]

{aria_code-0.112.0 → aria_code-0.113.0/src/aria_code.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aria-code
-Version: 0.112.0
+Version: 0.113.0
 Summary: Open-source AI agent for coding, financial analysis and logistics operations: numbers you can check, client data kept apart, actions a person approves
 License: Apache-2.0
 Project-URL: Homepage, https://github.com/artheras/aria-code
@@ -171,7 +171,7 @@ Numbers you can check, client data kept apart, actions a person approves.</p>
 <p align="center">
   <a href="https://www.npmjs.com/package/@artheras/aria-code"><img src="https://img.shields.io/npm/v/@artheras/aria-code/latest?style=flat-square&logo=npm&label=npm" alt="npm version"></a>
   <!-- Release automation pins PyPI: the default endpoint still selects legacy 4.x. -->
-  <a href="https://pypi.org/project/aria-code/0.112.0/"><img src="https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fpypi.org%2Fpypi%2Faria-code%2F0.112.0%2Fjson&amp;query=%24.info.version&amp;prefix=v&amp;style=flat-square&amp;logo=pypi&amp;label=PyPI&amp;color=blue&amp;cacheSeconds=300" alt="PyPI version"></a>
+  <a href="https://pypi.org/project/aria-code/0.113.0/"><img src="https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fpypi.org%2Fpypi%2Faria-code%2F0.113.0%2Fjson&amp;query=%24.info.version&amp;prefix=v&amp;style=flat-square&amp;logo=pypi&amp;label=PyPI&amp;color=blue&amp;cacheSeconds=300" alt="PyPI version"></a>
   <a href="https://github.com/artheras/aria-code/actions/workflows/ci.yml"><img src="https://img.shields.io/github/actions/workflow/status/artheras/aria-code/ci.yml?branch=main&style=flat-square&logo=githubactions&label=CI" alt="CI status"></a>
   <a href="LICENSE"><img src="https://img.shields.io/badge/license-Apache%202.0-64748b?style=flat-square" alt="Apache License 2.0"></a>
 </p>

{aria_code-0.112.0 → aria_code-0.113.0}/src/aria_code.egg-info/SOURCES.txt

@@ -459,6 +459,7 @@ src/aria_code/runtime/tool_executor.py
 src/aria_code/runtime/tool_policy.py
 src/aria_code/runtime/worktrees.py
 src/aria_code/safety/__init__.py
+src/aria_code/safety/client_isolation.py
 src/aria_code/safety/permissions.py
 src/aria_code/safety/sandbox.py
 src/aria_code/safety/service.py
@@ -548,6 +549,7 @@ tests/test_cli_lifecycle_hooks.py
 tests/test_cli_runtime_consumer.py
 tests/test_cli_tool_executor_module.py
 tests/test_cli_updater.py
+tests/test_client_isolation_guard.py
 tests/test_cloud_memory.py
 tests/test_code_audit_tools.py
 tests/test_code_review_agent.py
@@ -587,6 +589,7 @@ tests/test_engine_boundary.py
 tests/test_english_output_group.py
 tests/test_enterprise_tool_honesty.py
 tests/test_env_switch_docs.py
+tests/test_eval_hard_suite.py
 tests/test_eval_summary.py
 tests/test_eval_task_bank.py
 tests/test_eval_trajectories.py

aria_code-0.113.0/tests/test_client_isolation_guard.py

@@ -0,0 +1,102 @@
+"""A document written for one client cannot carry another client's data.
+
+gemini-3.5-flash, operations suite: report_BETA.md named Acme and Gamma and
+their SKUs, copied from notes on Beta's own rows. The output rules ask the
+model not to; this is the check that does not depend on it remembering.
+The data is the eval fixture itself: three shippers in one warehouse file.
+"""
+
+from __future__ import annotations
+
+import os
+import shutil
+from pathlib import Path
+
+import pytest
+
+from aria_code.apps.cli.tools.write_tools import tool_edit_file, tool_multi_edit, tool_write_file
+from aria_code.safety.client_isolation import check_write, client_index, foreign_mentions, target_client
+
+FIXTURE = Path(__file__).resolve().parents[1] / "evals" / "fixtures" / "shipper_isolation" / "inventory.csv"
+
+CLEAN = "# Beta Home Goods\n\nTotal units on hand: 560\n\n- BH-101: 25 on hand, safety stock 40\n"
+LEAKY = CLEAN + "- BH-106: 25 on hand — Re-labelled after GAMMA recall GP-104\n"
+
+
+@pytest.fixture
+def warehouse(tmp_path, monkeypatch):
+    shutil.copy(FIXTURE, tmp_path / "inventory.csv")
+    monkeypatch.chdir(tmp_path)
+    return tmp_path
+
+
+def test_clients_come_from_the_owner_column(warehouse):
+    index = client_index([warehouse])
+    assert set(index) == {"ACME", "BETA", "GAMMA"}
+    assert "Acme Outdoor Ltd" in index["ACME"].names
+    assert "AC-105" in index["ACME"].items
+
+
+def test_a_file_name_singles_out_one_client(warehouse):
+    index = client_index([warehouse])
+    assert target_client(Path("report_BETA.md"), index) == "BETA"
+    assert target_client(Path("beta-weekly.md"), index) == "BETA"
+    assert target_client(Path("BetaHomeGoods_stock.md"), index) == "BETA"
+    assert target_client(Path("warehouse_summary.md"), index) is None
+    assert target_client(Path("acme_vs_beta.md"), index) is None
+
+
+def test_other_clients_are_found_in_the_text(warehouse):
+    index = client_index([warehouse])
+    leaks = foreign_mentions("Shares rack R12 with Acme Outdoor Ltd AC-105. GAMMA recall GP-104.", "BETA", index)
+    assert set(leaks) == {"ACME", "GAMMA"}
+    assert "AC-105" in leaks["ACME"] and "GP-104" in leaks["GAMMA"]
+    assert foreign_mentions(CLEAN, "BETA", index) == {}
+
+
+def test_a_leaky_client_document_is_not_written(warehouse):
+    result = tool_write_file({"path": "report_BETA.md", "content": LEAKY})
+    assert not result["success"]
+    assert "GAMMA" in result["error"] and "GP-104" in result["error"]
+    assert not (warehouse / "report_BETA.md").exists()
+
+
+def test_a_clean_client_document_is_written(warehouse):
+    result = tool_write_file({"path": "report_BETA.md", "content": CLEAN})
+    assert result["success"], result
+    assert (warehouse / "report_BETA.md").read_text() == CLEAN
+
+
+def test_an_internal_document_may_name_every_client(warehouse):
+    result = tool_write_file({"path": "warehouse_summary.md", "content": LEAKY + "ACME 1169, GAMMA 1958\n"})
+    assert result["success"], result
+
+
+def test_code_is_not_checked(warehouse):
+    script = 'rows = [r for r in rows if r["owner_id"] == "BETA"]  # not ACME or GAMMA\n'
+    assert check_write(warehouse / "report_BETA.py", script) is None
+
+
+def test_an_edit_that_leaks_is_refused(warehouse):
+    (warehouse / "report_BETA.md").write_text(CLEAN)
+    result = tool_edit_file({"path": "report_BETA.md", "old_string": "Total units", "new_string": "Like ACME: total units"})
+    assert not result["success"] and "ACME" in result["error"]
+    assert (warehouse / "report_BETA.md").read_text() == CLEAN
+
+
+def test_a_multi_edit_that_leaks_is_refused(warehouse):
+    (warehouse / "report_BETA.md").write_text(CLEAN)
+    result = tool_multi_edit({"path": "report_BETA.md",
+                              "edits": [{"old_string": "560", "new_string": "560 (Gamma Pharma holds 1958)"}]})
+    assert not result["success"] and "GAMMA" in result["error"]
+
+
+def test_no_client_data_means_no_check(tmp_path, monkeypatch):
+    monkeypatch.chdir(tmp_path)
+    (tmp_path / "prices.csv").write_text("date,symbol,close\n2026-10-01,ACME,10\n")
+    assert check_write(tmp_path / "report_BETA.md", "ACME GAMMA") is None
+
+
+def test_a_single_client_file_is_not_a_multi_client_warehouse(tmp_path):
+    (tmp_path / "stock.csv").write_text("owner_id,sku,on_hand\nBETA,BH-1,4\n")
+    assert check_write(tmp_path / "report_BETA.md", "anything at all") is None