mantis-agent-sdk 2.1.0__tar.gz → 2.2.0__tar.gz

{mantis_agent_sdk-2.1.0 → mantis_agent_sdk-2.2.0}/CHANGELOG.md

@@ -74,6 +74,18 @@ The full versioning policy is in [SEMVER.md](SEMVER.md).
   Three new public exports: `ResponseFormatError`,
   `normalize_response_format`, `translate_response_format`.
 
+## [2.2.0] — 2026-06-30
+
+### Fixed
+
+- **`web_fetch` no longer depends on BeautifulSoup.** Its default (non-Exa) path
+  called `bs4` — not a dependency — so a plain `web_fetch(url)` returned raw HTML
+  (tags, `<script>`, CSS) as the model's "readable text". Rewritten with a
+  stdlib HTML→text extractor (drops script/style/head, block-closes → newlines,
+  strips tags, unescapes entities, collapses whitespace). Non-HTML bodies (JSON,
+  plain text, markdown, source) are now returned verbatim by content-type instead
+  of being tag-stripped. Same dependency-free treatment `web_search` got in 1.25.
+
 ## [2.1.0] — 2026-06-30
 
 ### Added

{mantis_agent_sdk-2.1.0 → mantis_agent_sdk-2.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mantis-agent-sdk
-Version: 2.1.0
+Version: 2.2.0
 Summary: Drop-in open-source agent SDK. Multi-model, streaming, MCP, sub-agents.
 Project-URL: Homepage, https://github.com/teddyoweh/mantis-agent-sdk
 Project-URL: Repository, https://github.com/teddyoweh/mantis-agent-sdk

{mantis_agent_sdk-2.1.0 → mantis_agent_sdk-2.2.0}/mantis_agent/__init__.py

@@ -284,7 +284,7 @@ def _detect_version() -> str:
 
         return version("mantis-agent-sdk")
     except Exception:  # pragma: no cover - extremely defensive
-        return "2.1.0"
+        return "2.2.0"
 
 
 __version__ = _detect_version()

{mantis_agent_sdk-2.1.0 → mantis_agent_sdk-2.2.0}/mantis_agent/builtin_tools/web.py

@@ -405,8 +405,35 @@ async def _exa_fetch(url: str, *, prompt: str | None) -> str:
     return "\n\n".join(parts) if parts else "(empty)"
 
 
+_DROP_BLOCK_RE = re.compile(
+    r"<(script|style|noscript|head|svg|template|iframe)\b[^>]*>.*?</\1\s*>",
+    re.DOTALL | re.IGNORECASE,
+)
+_BLOCK_CLOSE_RE = re.compile(
+    r"</(p|div|section|article|h[1-6]|li|tr|table|ul|ol|header|footer|nav|blockquote|pre)\s*>",
+    re.IGNORECASE,
+)
+_BR_RE = re.compile(r"<br\s*/?>", re.IGNORECASE)
+
+
+def _html_to_text(html: str) -> str:
+    """HTML → readable text, stdlib only (no BeautifulSoup). Drops script/style/
+    head blocks, turns block-element closes into newlines, strips remaining tags,
+    unescapes entities, and collapses blank runs."""
+    s = _DROP_BLOCK_RE.sub(" ", html)
+    s = _BR_RE.sub("\n", s)
+    s = _BLOCK_CLOSE_RE.sub("\n", s)
+    s = _TAG_RE.sub("", s)
+    s = html_mod.unescape(s)
+    lines = [re.sub(r"[ \t]+", " ", ln).strip() for ln in s.splitlines()]
+    return "\n".join(ln for ln in lines if ln)
+
+
 async def _raw_fetch(url: str) -> str:
-    """Direct HTTP fetch + BeautifulSoup text extraction. Fallback path."""
+    """Direct HTTP fetch + dependency-free text extraction (the default path when
+    Exa isn't configured). HTML is cleaned to readable text; non-HTML bodies
+    (JSON, plain text, markdown, source) are returned verbatim so nothing useful
+    is mangled."""
 
     try:
         r = await _client().get(url)
@@ -414,17 +441,13 @@ async def _raw_fetch(url: str) -> str:
     except httpx.HTTPError as e:
         return f"fetch error: {e!r}"
 
-    try:
-        from bs4 import BeautifulSoup  # type: ignore[import-untyped]
-    except ImportError:
-        # No bs4 — return raw text truncated.
-        return r.text[:5000]
-
-    soup = BeautifulSoup(r.text, "html.parser")
-    for el in soup(["script", "style", "noscript"]):
-        el.decompose()
-    text = " ".join(soup.get_text(" ", strip=True).split())
-    return text[:5000] if text else "(empty page)"
+    ctype = r.headers.get("content-type", "").lower()
+    body = r.text
+    is_html = "html" in ctype or (not ctype and "<html" in body[:2000].lower())
+    if is_html:
+        text = _html_to_text(body)
+        return text[:5000] if text else "(empty page)"
+    return body[:5000] if body else "(empty)"
 
 
 # ---------------------------------------------------------------------------

{mantis_agent_sdk-2.1.0 → mantis_agent_sdk-2.2.0}/mantis_agent/catalog.py

@@ -200,11 +200,18 @@ def set_key(provider_id: str, key: str) -> None:
 def clear_key(provider_id: str) -> bool:
     data = _load_store()
     keys = data.get("keys") or {}
-    if provider_id in keys:
+    removed = provider_id in keys
+    if removed:
         del keys[provider_id]
         _save_store(data)
-        return True
-    return False
+    # Mirror set_key, which also exports the key into the process env — otherwise
+    # api_key_for() would still find it there and the provider would stay
+    # "enabled" after a /disable. (A shell-exported var reappears next launch,
+    # which is the right behavior — that's the user's standing choice.)
+    prov = BY_ID.get(provider_id)
+    if prov and prov.api_key_env:
+        os.environ.pop(prov.api_key_env, None)
+    return removed
 
 
 def get_last_model() -> dict[str, Any] | None:

{mantis_agent_sdk-2.1.0 → mantis_agent_sdk-2.2.0}/mantis_agent/tui_fullscreen.py

@@ -451,6 +451,13 @@ async def run_fullscreen(tui: Any) -> int:
             await _announce(f"✗ {prov.label}: {detail} (key not saved)")
             return
         _switch_model(model, provider_id=pid)
+        # Fetch the provider's real /v1/models now (off-thread) so the next
+        # /models shows its full catalog, not just the flagship starter list.
+        try:
+            await asyncio.to_thread(catalog.refresh_live_models, prov)
+            state.pop("model_cache", None)  # invalidate the active-backend cache
+        except Exception:  # noqa: BLE001
+            pass
         await _announce(f"✓ {prov.label} enabled · model → {model}")
 
     def _width() -> int:

{mantis_agent_sdk-2.1.0 → mantis_agent_sdk-2.2.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "mantis-agent-sdk"
-version = "2.1.0"
+version = "2.2.0"
 description = "Drop-in open-source agent SDK. Multi-model, streaming, MCP, sub-agents."
 readme = "README.md"
 requires-python = ">=3.11"

{mantis_agent_sdk-2.1.0 → mantis_agent_sdk-2.2.0}/tests/test_model_setup_sources.py

@@ -127,6 +127,23 @@ def test_grouped_provider_models_enabled_before_disabled() -> None:
             assert not seen_disabled, "an enabled group appeared after a disabled one"
 
 
+def test_setting_a_key_enables_and_regroups_provider(monkeypatch, tmp_path) -> None:
+    # The inline-enable path relies on this: set_key → is_enabled True → the
+    # provider is grouped under "enabled". clear_key reverses it.
+    monkeypatch.setenv("MANTIS_AGENT_HOME", str(tmp_path))
+    monkeypatch.delenv("DEEPSEEK_API_KEY", raising=False)
+    prov = catalog.BY_ID["deepseek"]
+    assert catalog.is_enabled(prov) is False
+    catalog.set_key("deepseek", "sk-test-key")
+    try:
+        assert catalog.is_enabled(prov) is True
+        enabled_ids = [g["provider_id"] for g in catalog.grouped_provider_models() if g["enabled"]]
+        assert "deepseek" in enabled_ids
+    finally:
+        catalog.clear_key("deepseek")
+    assert catalog.is_enabled(prov) is False
+
+
 def test_grouped_provider_models_are_real_providers_with_lists() -> None:
     for g in catalog.grouped_provider_models():
         assert g["provider_id"] in catalog.BY_ID

mantis_agent_sdk-2.2.0/tests/test_web_fetch.py

@@ -0,0 +1,62 @@
+"""web_fetch — dependency-free HTML→text extraction (no BeautifulSoup)."""
+
+from __future__ import annotations
+
+import anyio
+import httpx
+
+import mantis_agent.builtin_tools.web as web
+from mantis_agent.builtin_tools.web import _html_to_text, web_fetch
+
+
+def test_html_to_text_drops_script_style() -> None:
+    html = ("<html><head><style>.x{color:red}</style></head><body>"
+            "<script>var secret=1;</script><h1>Hi &amp; bye</h1>"
+            "<p>Para one.</p><p>Para two.</p></body></html>")
+    out = _html_to_text(html)
+    assert "var secret" not in out and "color:red" not in out
+    assert "Hi & bye" in out                      # entity decoded
+    assert "Para one." in out and "Para two." in out
+
+
+def test_html_to_text_block_newlines() -> None:
+    out = _html_to_text("<p>a</p><p>b</p><br><div>c</div>")
+    assert out.splitlines() == ["a", "b", "c"]
+
+
+def test_html_to_text_collapses_whitespace() -> None:
+    assert _html_to_text("<p>lots     of    space</p>") == "lots of space"
+
+
+def _mock_web(monkeypatch, *, body: str, content_type: str) -> None:
+    def handler(request: httpx.Request) -> httpx.Response:
+        return httpx.Response(200, text=body, headers={"content-type": content_type})
+    client = httpx.AsyncClient(transport=httpx.MockTransport(handler))
+    monkeypatch.setattr(web, "_CLIENT", client)
+    monkeypatch.delenv("EXA_API_KEY", raising=False)   # force the raw path
+
+
+def test_web_fetch_html_cleaned(monkeypatch) -> None:
+    _mock_web(monkeypatch,
+              body="<html><body><script>x</script><p>Clean me</p></body></html>",
+              content_type="text/html; charset=utf-8")
+    out = anyio.run(lambda: web_fetch.fn(url="http://x"))
+    assert "Clean me" in out
+    assert "<p>" not in out and "script" not in out
+
+
+def test_web_fetch_json_verbatim(monkeypatch) -> None:
+    _mock_web(monkeypatch, body='{"key": "value", "n": 1}',
+              content_type="application/json")
+    out = anyio.run(lambda: web_fetch.fn(url="http://x/api"))
+    assert out == '{"key": "value", "n": 1}'         # NOT tag-stripped
+
+
+def test_web_fetch_http_error(monkeypatch) -> None:
+    def handler(request: httpx.Request) -> httpx.Response:
+        return httpx.Response(404, text="nope")
+    client = httpx.AsyncClient(transport=httpx.MockTransport(handler))
+    monkeypatch.setattr(web, "_CLIENT", client)
+    monkeypatch.delenv("EXA_API_KEY", raising=False)
+    out = anyio.run(lambda: web_fetch.fn(url="http://x"))
+    assert "fetch error" in out