brainpick 0.3.2__tar.gz → 0.4.0__tar.gz

{brainpick-0.3.2 → brainpick-0.4.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: brainpick
-Version: 0.3.2
+Version: 0.4.0
 Summary: A turn-key brain stack for agents: compile an OKF markdown bundle into a queryable, servable, visualizable brain.
 Author: Benquemax
 License: MIT

{brainpick-0.3.2 → brainpick-0.4.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "brainpick"
-version = "0.3.2"
+version = "0.4.0"
 description = "A turn-key brain stack for agents: compile an OKF markdown bundle into a queryable, servable, visualizable brain."
 readme = "README.md"
 requires-python = ">=3.10"

{brainpick-0.3.2 → brainpick-0.4.0}/src/brainpick/__init__.py

@@ -1,4 +1,4 @@
 """brainpick — a turn-key brain stack for agents."""
 
-__version__ = "0.3.2"
+__version__ = "0.4.0"
 SPEC_VERSION = "0.1"

{brainpick-0.3.2 → brainpick-0.4.0}/src/brainpick/cli.py

@@ -500,7 +500,7 @@ def build_parser() -> argparse.ArgumentParser:
 
     p_integrate = sub.add_parser("integrate", help="install brainpick into an agent harness (skill, MCP, report)")
     p_integrate.add_argument("target", metavar="<target>",
-                             help="the harness to wire up: claude-code | opencode | agents-md")
+                             help="the harness to wire up: claude-code | opencode | agents-md | dsh")
     p_integrate.add_argument("--root", default=".", help="bundle root (default: current directory)")
     p_integrate.add_argument("--dry-run", action="store_true",
                              help="print what integrate would do without writing anything")

{brainpick-0.3.2 → brainpick-0.4.0}/src/brainpick/integrate.py

@@ -9,6 +9,9 @@ Three targets, one family voice (mirrors henxels' `integrate`):
   opencode.json MCP snippet.
 - `agents-md`    — ensure an AGENTS.md exists (the one place integrate may create a
   file), install the brain-report markers if absent, and compile so the block fills.
+- `dsh`          — write the Agent Skill (the `.claude/skills` convention dsh shares),
+  then PRINT the `cordis.patch.yml` insert that mounts brainpick through
+  `@deepseek-ai/dsh-mcp-client` (dsh has no `claude mcp add`; a server is a config row).
 
 The shipped Agent Skill (integrations/skill/SKILL.md, canonical) rides inside the
 package; the parity test asserts the shipped copy is byte-identical to the canonical.
@@ -22,15 +25,17 @@ from pathlib import Path
 from brainpick.compile.pipeline import run_compile
 from brainpick.compile.t1 import REPORT_BEGIN_PREFIX, REPORT_END_MARKER
 from brainpick.detect import find_repo_root
-from brainpick.scaffold import _Voice, mcp_snippets
+from brainpick.scaffold import _Voice, dsh_snippet, mcp_snippets
 
-TARGETS = ("claude-code", "opencode", "agents-md")
+TARGETS = ("claude-code", "opencode", "agents-md", "dsh")
 HENXELS_BEGIN = "<!-- henxels:begin -->"
 
 # harness -> where its Agent Skill lands, relative to the repo root
 SKILL_DESTINATIONS = {
     "claude-code": Path(".claude") / "skills" / "brainpick" / "SKILL.md",
     "opencode": Path(".opencode") / "skills" / "brainpick" / "SKILL.md",
+    # dsh loads Agent Skills from the same .claude/skills convention as Claude Code.
+    "dsh": Path(".claude") / "skills" / "brainpick" / "SKILL.md",
 }
 
 _MINIMAL_AGENTS = "# AGENTS.md\n\nWorking notes for agents in this repository.\n"
@@ -128,6 +133,18 @@ def _integrate_opencode(voice: _Voice, root: Path, repo: Path, dry_run: bool) ->
     return 0
 
 
+def _integrate_dsh(voice: _Voice, root: Path, repo: Path, dry_run: bool) -> int:
+    dest, existed = _write_skill(repo, "dsh", dry_run)
+    verb = "would write" if dry_run else ("updated" if existed else "wrote")
+    voice.line("✓", f"skill: {verb} {dest}")
+    if dry_run:
+        voice.step("• print the cordis.patch.yml insert for @deepseek-ai/dsh-mcp-client")
+        return 0
+    voice.raw()
+    voice.raw(dsh_snippet(root))
+    return 0
+
+
 def _integrate_agents_md(voice: _Voice, root: Path, repo: Path, dry_run: bool) -> int:
     agents = repo / "AGENTS.md"
     existed = agents.is_file()
@@ -177,4 +194,6 @@ def run_integrate(target: str, root: str | Path, dry_run: bool = False) -> int:
         return _integrate_claude_code(voice, root, repo, dry_run)
     if target == "opencode":
         return _integrate_opencode(voice, root, repo, dry_run)
+    if target == "dsh":
+        return _integrate_dsh(voice, root, repo, dry_run)
     return _integrate_agents_md(voice, root, repo, dry_run)

{brainpick-0.3.2 → brainpick-0.4.0}/src/brainpick/mcp_server.py

@@ -26,6 +26,7 @@ from brainpick.federation import (
 )
 from brainpick.llm import make_chat
 from brainpick.merge import find_base, resolve
+from brainpick.query.keyword import tokenize
 from brainpick.query.router import KNOWN_MODES, run_search
 from brainpick.serve.state import ServeState, bfs_neighborhood, jsonable, resolve_doc
 from brainpick.serve.watcher import recompile_and_broadcast
@@ -106,6 +107,25 @@ def _single_overview(state: ServeState, budget_tokens: int | None = None) -> dic
 # -- brain_search ------------------------------------------------------------------
 
 
+def _matched_terms(query: str, *fields: str) -> list[str]:
+    """The query tokens that actually occur in the given fields, in query order,
+    de-duplicated. Uses the same tokenizer keyword search uses, so 'why' reports
+    what really matched instead of echoing the whole query string (issue #2)."""
+    field_tokens = set()
+    for field in fields:
+        if field:
+            field_tokens.update(tokenize(field))
+    seen: dict[str, None] = {}
+    for token in tokenize(query):
+        if token in field_tokens and token not in seen:
+            seen[token] = None
+    return list(seen)
+
+
+def _quote_terms(terms: list[str]) -> str:
+    return ", ".join(f"'{t}'" for t in terms)
+
+
 def _why(hit: dict, query: str) -> str:
     lowered = query.lower()
     if lowered in str(hit["title"]).lower():
@@ -116,7 +136,20 @@ def _why(hit: dict, query: str) -> str:
         return f"semantically close to '{query}'"
     if hit.get("source") == "graph":
         return f"connected in the entity graph to '{query}'"
-    return f"body mentions '{query}'" if hit.get("snippet") else "keyword match"
+    # Keyword-ish hit: name the query tokens that actually occur, and where, rather
+    # than claiming the whole query appears verbatim (issue #2).
+    total = len(dict.fromkeys(tokenize(query)))
+    title_terms = _matched_terms(query, str(hit.get("title") or ""))
+    desc_terms = _matched_terms(query, hit.get("description") or "")
+    body_terms = _matched_terms(query, hit.get("snippet") or "")
+    if title_terms:
+        return f"title mentions {_quote_terms(title_terms)}"
+    if desc_terms:
+        return f"description mentions {_quote_terms(desc_terms)}"
+    if body_terms:
+        suffix = f" ({len(body_terms)}/{total} terms)" if total > 1 else ""
+        return f"body mentions {_quote_terms(body_terms)}{suffix}"
+    return "keyword match"
 
 
 def _single_search(state: ServeState, query: str, mode: str = "auto", limit: int = 8,

{brainpick-0.3.2 → brainpick-0.4.0}/src/brainpick/scaffold.py

@@ -212,6 +212,38 @@ def mcp_snippets(bundle: Path) -> str:
     return "\n".join(parts)
 
 
+def dsh_snippet(bundle: Path) -> str:
+    """The DeepSeek Harness (dsh) wiring: an insert entry for the web profile's
+    `cordis.patch.yml` that mounts brainpick through `@deepseek-ai/dsh-mcp-client`
+    as a stdio server. dsh has no `claude mcp add` equivalent — a server is a
+    config row, so we print the exact row with the resolved absolute root."""
+    command = brainpick_command() + ["mcp", "--root", str(bundle)]
+    entry = (
+        "    - id: mcp-brainpick\n"
+        "      name: '@deepseek-ai/dsh-mcp-client'\n"
+        "      config:\n"
+        "        serverName: brainpick\n"
+        "        transport: stdio\n"
+        f"        command: {command[0]}\n"
+        f"        args: {json.dumps(command[1:])}\n"
+        "        failOnStartupError: false"
+    )
+    parts = [
+        "Wire brainpick into DeepSeek Harness (dsh):",
+        "",
+        "  Add this insert entry to your web profile's cordis.patch.yml",
+        "  (~/.dsh/profiles/web/cordis.patch.yml), then restart the service",
+        "  (e.g. systemctl --user restart dsh.service) and hard-refresh:",
+        "",
+        "  - insert:",
+        entry,
+        "",
+        "  Tools then surface as mcp__brainpick__brain_* — brain_overview first,",
+        "  then brain_search / brain_read / brain_neighbors.",
+    ]
+    return "\n".join(parts)
+
+
 def henxels_fragment(contract: Path, bundle: Path) -> str:
     """The freshness gate, paste-able into an existing contract — never applied for you."""
     root = os.path.relpath(bundle, contract.parent)

{brainpick-0.3.2 → brainpick-0.4.0}/src/brainpick.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: brainpick
-Version: 0.3.2
+Version: 0.4.0
 Summary: A turn-key brain stack for agents: compile an OKF markdown bundle into a queryable, servable, visualizable brain.
 Author: Benquemax
 License: MIT

{brainpick-0.3.2 → brainpick-0.4.0}/tests/test_e2e_serve.py

@@ -107,7 +107,7 @@ def test_sse_frame_format():
 def test_health(kotiaurinko):
     with TestClient(make_app(kotiaurinko)) as client:
         assert client.get("/api/health").json() == {
-            "impl": "python", "name": "brainpick", "spec_version": "0.1", "version": "0.3.2",
+            "impl": "python", "name": "brainpick", "spec_version": "0.1", "version": "0.4.0",
         }
 
 

{brainpick-0.3.2 → brainpick-0.4.0}/tests/test_integrate.py

@@ -53,10 +53,25 @@ def test_integrate_opencode_writes_skill_under_its_convention(repo, capsys):
     assert "opencode" in capsys.readouterr().out.lower()
 
 
+def test_integrate_dsh_writes_skill_and_prints_cordis_insert(repo, capsys):
+    root, bundle = repo
+    assert run_integrate("dsh", bundle) == 0
+    skill = root / SKILL_DESTINATIONS["dsh"]
+    assert skill.read_text(encoding="utf-8") == CANONICAL.read_text(encoding="utf-8")
+    out = capsys.readouterr().out
+    assert "@deepseek-ai/dsh-mcp-client" in out       # the mcp-client bundle
+    assert "cordis.patch.yml" in out                   # where the row goes
+    assert "serverName: brainpick" in out
+    assert str(bundle) in out                          # resolved absolute root, not a stale path
+    assert "claude mcp add" not in out                 # dsh has no such command
+
+
 def test_integrate_dry_run_is_inert(repo, capsys):
     root, bundle = repo
     assert run_integrate("claude-code", bundle, dry_run=True) == 0
     assert not (root / ".claude").exists()
+    assert run_integrate("dsh", bundle, dry_run=True) == 0
+    assert not (root / ".claude").exists()
     assert run_integrate("agents-md", bundle, dry_run=True) == 0
     assert not (root / "AGENTS.md").exists()
 

{brainpick-0.3.2 → brainpick-0.4.0}/tests/test_mcp_tools.py

@@ -102,6 +102,22 @@ def test_search_hits_have_why_not_bodies(kotiaurinko):
     assert result["truncated"] is False
 
 
+def test_search_why_names_matched_terms_not_whole_query(kotiaurinko):
+    """issue #2: a multi-word query must not claim the WHOLE query appears in a
+    doc that only matched some of its terms. The 'why' names the tokens that
+    actually occur, never a term (like 'banana') that appears nowhere."""
+    result = search_payload(make_state(kotiaurinko), "aurinko komeetta banana")
+    assert result["hits"], "expected keyword hits for a partially-matching query"
+    for hit in result["hits"]:
+        why = hit["why"]
+        # never echo the full query, and never claim a non-existent term matched
+        assert "aurinko komeetta banana" not in why
+        assert "banana" not in why
+        # a keyword-ish 'mentions' reason must quote the actual matched token(s)
+        if "mentions" in why:
+            assert "aurinko" in why or "komeetta" in why
+
+
 def test_search_budget_trims_hits(kotiaurinko):
     state = make_state(kotiaurinko)
     full = search_payload(state, "aurinko")