specsmith 0.16.5.dev630__tar.gz → 0.16.5.dev640__tar.gz

{specsmith-0.16.5.dev630/src/specsmith.egg-info → specsmith-0.16.5.dev640}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: specsmith
-Version: 0.16.5.dev630
+Version: 0.16.5.dev640
 Summary: AEE governance toolkit for AI-assisted development — session preflight gates, multi-agent dispatch, requirements↔test traceability, ESDB persistence, MCP server, and skills for Warp, Cursor, Claude Code, Copilot, Windsurf, and Aider.
 Author: Layer1Labs Silicon, Inc.
 License-Expression: MIT
@@ -261,11 +261,16 @@ command — not to library imports.  Use this when you want the AEE belief-state
 in your own application without managing a pipx environment.
 
 **ESDB — Epistemic State Database**
+Terminology used in this repo is strict:
+- **ESDB** = the specification/data model category.
+- **SQLite backend** = the free/default ESDB implementation bundled in `specsmith`.
+- **ChronoMemory** = the commercial package.
+- **ChronoStore** = the backend engine/class provided by the ChronoMemory package.
 
 | Tier | Package | License | What you get |
 |------|---------|---------|-------------|
 | **Default** | `specsmith` (built-in) | MIT, free | SQLite backend — requirements, test cases, confidence filtering |
-| **Commercial** | `chronomemory` via `specsmith[esdb]` | Proprietary — license required (see [COMMERCIAL-LICENSE.md](https://github.com/layer1labs/specsmith/blob/develop/COMMERCIAL-LICENSE.md)) | ChronoStore: tamper-evident SHA-256 WAL, OEA anti-hallucination fields, Rust acceleration, epistemic rollback |
+| **Commercial** | `chronomemory` via `specsmith[esdb]` | Proprietary — license required (see [COMMERCIAL-LICENSE.md](https://github.com/layer1labs/specsmith/blob/develop/COMMERCIAL-LICENSE.md)) | ChronoMemory package (ChronoStore backend): tamper-evident SHA-256 WAL, OEA anti-hallucination fields, Rust acceleration, epistemic rollback |
 
 See `docs/editions.md` for the full OSS vs commercial feature matrix.
 
@@ -281,7 +286,7 @@ specsmith esdb status   # shows: SQLite (free, MIT) — active by default
 - Commit canonical ChronoMemory state files: `.chronomemory/events.wal` and `.chronomemory/snapshot.json`.
 - Do not commit ChronoMemory timestamped backup copies under `.chronomemory/backup/` (regenerated by `specsmith save` / `specsmith esdb backup`).
 
-**Upgrading to chronomemory ChronoStore (commercial):**
+**Upgrading to ChronoMemory (ChronoStore backend, commercial):**
 
 If you hold a chronomemory ESDB license, activate the commercial backend:
 

{specsmith-0.16.5.dev630 → specsmith-0.16.5.dev640}/README.md

@@ -183,11 +183,16 @@ command — not to library imports.  Use this when you want the AEE belief-state
 in your own application without managing a pipx environment.
 
 **ESDB — Epistemic State Database**
+Terminology used in this repo is strict:
+- **ESDB** = the specification/data model category.
+- **SQLite backend** = the free/default ESDB implementation bundled in `specsmith`.
+- **ChronoMemory** = the commercial package.
+- **ChronoStore** = the backend engine/class provided by the ChronoMemory package.
 
 | Tier | Package | License | What you get |
 |------|---------|---------|-------------|
 | **Default** | `specsmith` (built-in) | MIT, free | SQLite backend — requirements, test cases, confidence filtering |
-| **Commercial** | `chronomemory` via `specsmith[esdb]` | Proprietary — license required (see [COMMERCIAL-LICENSE.md](https://github.com/layer1labs/specsmith/blob/develop/COMMERCIAL-LICENSE.md)) | ChronoStore: tamper-evident SHA-256 WAL, OEA anti-hallucination fields, Rust acceleration, epistemic rollback |
+| **Commercial** | `chronomemory` via `specsmith[esdb]` | Proprietary — license required (see [COMMERCIAL-LICENSE.md](https://github.com/layer1labs/specsmith/blob/develop/COMMERCIAL-LICENSE.md)) | ChronoMemory package (ChronoStore backend): tamper-evident SHA-256 WAL, OEA anti-hallucination fields, Rust acceleration, epistemic rollback |
 
 See `docs/editions.md` for the full OSS vs commercial feature matrix.
 
@@ -203,7 +208,7 @@ specsmith esdb status   # shows: SQLite (free, MIT) — active by default
 - Commit canonical ChronoMemory state files: `.chronomemory/events.wal` and `.chronomemory/snapshot.json`.
 - Do not commit ChronoMemory timestamped backup copies under `.chronomemory/backup/` (regenerated by `specsmith save` / `specsmith esdb backup`).
 
-**Upgrading to chronomemory ChronoStore (commercial):**
+**Upgrading to ChronoMemory (ChronoStore backend, commercial):**
 
 If you hold a chronomemory ESDB license, activate the commercial backend:
 

{specsmith-0.16.5.dev630 → specsmith-0.16.5.dev640}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "specsmith"
-version = "0.16.5.dev630"
+version = "0.16.5.dev640"
 description = "AEE governance toolkit for AI-assisted development — session preflight gates, multi-agent dispatch, requirements↔test traceability, ESDB persistence, MCP server, and skills for Warp, Cursor, Claude Code, Copilot, Windsurf, and Aider."
 readme = "README.md"
 license = "MIT"
@@ -182,6 +182,8 @@ module = [
     "whisper_cpp_python.*",
     "chronomemory",   # git dep — no published stubs yet
     "chronomemory.*",
+    "tomllib",         # stdlib in 3.11+; not available when checking against python_version=3.10
+    "tomli",           # optional 3.10 backport; not installed in the dev environment
 ]
 ignore_missing_imports = true
 

{specsmith-0.16.5.dev630 → specsmith-0.16.5.dev640}/src/specsmith/agent/broker.py

@@ -71,6 +71,7 @@ class Intent(str, Enum):
 
     READ_ONLY_ASK = "read_only_ask"
     CHANGE = "change"
+    REFACTOR = "refactor"  # structural change with no behaviour change — highest scope-creep risk
     RELEASE = "release"
     DESTRUCTIVE = "destructive"
 
@@ -95,10 +96,19 @@ _RELEASE_PATTERNS = (
     re.compile(r"\bbump\s+(the\s+)?version\b", re.IGNORECASE),
 )
 
+_REFACTOR_PATTERNS = (
+    re.compile(r"\b(refactor|rewrite|extract|reorganise|reorganize|restructure)\b", re.IGNORECASE),
+    re.compile(r"\brename\s+\w", re.IGNORECASE),  # rename <identifier>
+    re.compile(r"\bmove\s+.+?\bto\b", re.IGNORECASE),  # move X to Y
+    re.compile(r"\bsplit\s+.+?\binto\b", re.IGNORECASE),  # split X into Y
+    re.compile(r"\bconsolidate\b", re.IGNORECASE),
+    re.compile(r"\binline\s+the\b", re.IGNORECASE),
+)
+
 _CHANGE_PATTERNS = (
     re.compile(r"\b(fix|repair|patch)\b", re.IGNORECASE),
     re.compile(r"\b(add|implement|create|introduce|build)\b", re.IGNORECASE),
-    re.compile(r"\b(refactor|rewrite|rename|extract)\b", re.IGNORECASE),
+    re.compile(r"\b(rename)\b", re.IGNORECASE),
     re.compile(r"\b(update|migrate|upgrade)\b", re.IGNORECASE),
     re.compile(r"\b(remove|delete)\s+(the\s+)?(unused|stale|legacy)\b", re.IGNORECASE),
 )
@@ -113,8 +123,10 @@ _READ_ONLY_PATTERNS = (
 def classify_intent(utterance: str) -> Intent:
     """Classify the user's natural-language utterance.
 
-    Order: DESTRUCTIVE > RELEASE > CHANGE > READ_ONLY_ASK > READ_ONLY_ASK
-    (default).
+    Order: DESTRUCTIVE > RELEASE > REFACTOR > CHANGE > READ_ONLY_ASK.
+    REFACTOR is checked before CHANGE so that structural-change verbs
+    (refactor, extract, reorganise ...) get the tighter scope-discipline
+    governance rather than the generic change governance.
     """
     if not utterance or not utterance.strip():
         return Intent.READ_ONLY_ASK
@@ -125,6 +137,9 @@ def classify_intent(utterance: str) -> Intent:
     for p in _RELEASE_PATTERNS:
         if p.search(text):
             return Intent.RELEASE
+    for p in _REFACTOR_PATTERNS:
+        if p.search(text):
+            return Intent.REFACTOR
     for p in _CHANGE_PATTERNS:
         if p.search(text):
             return Intent.CHANGE

specsmith-0.16.5.dev640/src/specsmith/agent/model_router.py

@@ -0,0 +1,242 @@
+# SPDX-License-Identifier: MIT
+# Copyright (c) 2026 Layer1Labs Silicon, Inc. All rights reserved.
+"""Per-turn intent classifier and Ollama model router (REQ-388, REQ-389).
+
+``ModelRouter`` classifies each user utterance into one of three roles
+(``general``, ``coding``, ``reasoning``) using a fast keyword + slash-command
+heuristic and returns the best Ollama model tag for that role.  The router
+only changes the active model when the role classification changes, so a
+long coding session stays on the coder model without emitting a switch event
+on every turn.
+
+Usage
+-----
+::
+
+    router = ModelRouter({"general": "qwen2.5:14b",
+                          "coding": "qwen2.5-coder:14b",
+                          "reasoning": "deepseek-r1:14b"})
+    model_tag, switched = router.route("write a function that ...")
+    if switched:
+        print(f"→ switched to {model_tag}")
+"""
+
+from __future__ import annotations
+
+__all__ = ["ModelRouter", "classify_intent"]
+
+# ---------------------------------------------------------------------------
+# Intent classification
+# ---------------------------------------------------------------------------
+
+# Slash commands that strongly signal coding work.
+_CODING_SLASH: frozenset[str] = frozenset({"/code", "/fix", "/refactor", "/test", "/commit", "/pr"})
+
+# Slash commands that signal analytical / planning work.
+_REASONING_SLASH: frozenset[str] = frozenset({"/architect", "/plan", "/audit", "/review", "/why"})
+
+# Keywords that indicate the user wants code written or bugs fixed.
+_CODING_KEYWORDS: frozenset[str] = frozenset(
+    {
+        "write",
+        "implement",
+        "function",
+        "method",
+        "class",
+        "bug",
+        "error",
+        "fix",
+        "patch",
+        "test",
+        "debug",
+        "refactor",
+        "import",
+        "syntax",
+        "compile",
+        "code",
+        "script",
+        "module",
+        "api",
+        "endpoint",
+        "snippet",
+        "variable",
+        "loop",
+        "array",
+        "dict",
+        "parse",
+        "format",
+        "regex",
+        "unittest",
+        "pytest",
+        "mock",
+        "stub",
+        "boilerplate",
+        "scaffold",
+        "decorator",
+        "lambda",
+        "iterator",
+        "generator",
+        "async",
+        "await",
+        "coroutine",
+        "type",
+        "hint",
+        "annotation",
+        "dataclass",
+    }
+)
+
+# Keywords that indicate the user wants analysis, architecture, or deep reasoning.
+_REASONING_KEYWORDS: frozenset[str] = frozenset(
+    {
+        "analyze",
+        "analyse",
+        "explain",
+        "reason",
+        "strategy",
+        "architecture",
+        "design",
+        "evaluate",
+        "compare",
+        "tradeoff",
+        "trade-off",
+        "review",
+        "audit",
+        "assess",
+        "consider",
+        "approach",
+        "decision",
+        "structure",
+        "plan",
+        "requirements",
+        "spec",
+        "diagram",
+        "model",
+        "rationale",
+        "justify",
+        "pros",
+        "cons",
+        "recommend",
+        "alternative",
+        "consequence",
+        "impact",
+        "risk",
+        "bottleneck",
+        "complexity",
+        "scalability",
+    }
+)
+
+
+def classify_intent(text: str) -> str:
+    """Classify *text* as ``"coding"``, ``"reasoning"``, or ``"general"``.
+
+    The classification is intentionally fast (O(n) word scan, no ML) so it
+    adds negligible latency to each turn.  Rules in priority order:
+
+    1. Slash-command prefix → maps directly to a role.
+    2. Keyword match → first-match wins; coding keywords checked before
+       reasoning keywords so ``/fix a bug in the design docs`` routes to
+       coding.
+    3. Default → ``"general"``.
+    """
+    text_stripped = text.strip()
+
+    # --- Rule 1: explicit slash-command ---
+    head = text_stripped.split()[0].lower() if text_stripped else ""
+    if head in _CODING_SLASH:
+        return "coding"
+    if head in _REASONING_SLASH:
+        return "reasoning"
+
+    # --- Rule 2: keyword scan ---
+    words = set(text_stripped.lower().split())
+    if words & _CODING_KEYWORDS:
+        return "coding"
+    if words & _REASONING_KEYWORDS:
+        return "reasoning"
+
+    return "general"
+
+
+# ---------------------------------------------------------------------------
+# ModelRouter
+# ---------------------------------------------------------------------------
+
+
+class ModelRouter:
+    """Routes each user turn to the appropriate Ollama model (REQ-388).
+
+    Parameters
+    ----------
+    roles:
+        Mapping of role name (``"general"``, ``"coding"``, ``"reasoning"``)
+        to Ollama model tag.  Missing roles fall back to ``"general"``; if
+        ``"general"`` is also missing the router returns ``(None, False)``
+        and the caller uses whatever model was already set.
+    """
+
+    def __init__(self, roles: dict[str, str | None]) -> None:
+        # Strip out None values so lookups never return None.
+        self._roles: dict[str, str] = {k: v for k, v in roles.items() if v}
+        self._current_role: str | None = None
+        self._current_model: str | None = None
+
+    # ── Public interface ────────────────────────────────────────────────────
+
+    def route(self, text: str) -> tuple[str | None, bool]:
+        """Classify *text* and return the model tag for the detected role.
+
+        Returns
+        -------
+        (model_tag, switched)
+            *model_tag* is the Ollama tag to use for this turn, or ``None``
+            when no role→model mapping is configured.
+            *switched* is ``True`` when the model changed from the previous
+            turn (useful for emitting a switch notification).
+        """
+        if not self._roles:
+            return None, False
+
+        role = classify_intent(text)
+
+        # Resolve: prefer exact role match, fallback to general, then any.
+        model = (
+            self._roles.get(role) or self._roles.get("general") or next(iter(self._roles.values()))
+        )
+
+        if model == self._current_model:
+            return model, False
+
+        prev_role = self._current_role
+        self._current_role = role
+        self._current_model = model
+        switched = prev_role is not None  # first call is not a "switch"
+        return model, switched
+
+    @property
+    def active_model(self) -> str | None:
+        """Model tag currently in use, or ``None`` before the first turn."""
+        return self._current_model
+
+    @property
+    def active_role(self) -> str | None:
+        """Role name currently in use, or ``None`` before the first turn."""
+        return self._current_role
+
+    def table(self) -> str:
+        """Human-readable routing table for the ``/models`` slash command."""
+        if not self._roles:
+            return "  (no multi-model routing configured)"
+        role_order = ["general", "coding", "reasoning"]
+        lines: list[str] = ["  Multi-model routing:"]
+        for role in role_order:
+            if role in self._roles:
+                active_marker = " ◀ active" if role == self._current_role else ""
+                lines.append(f"    {role:<10} → {self._roles[role]}{active_marker}")
+        # Any roles not in the standard order
+        for role, model in self._roles.items():
+            if role not in role_order:
+                active_marker = " ◀ active" if role == self._current_role else ""
+                lines.append(f"    {role:<10} → {model}{active_marker}")
+        return "\n".join(lines)

{specsmith-0.16.5.dev630 → specsmith-0.16.5.dev640}/src/specsmith/agent/runner.py

@@ -237,6 +237,7 @@ SLASH_COMMANDS: dict[str, str] = {
     "/undo": "[UNDO] Revert the last action: ",
     "/context": "[CONTEXT] Surface repo context relevant to: ",
     "/search": "[SEARCH] Search the repo and external docs for: ",
+    "/models": "",  # handled inline — shows the multi-model routing table
 }
 
 
@@ -334,6 +335,10 @@ class AgentRunner:
         # back to single-profile behaviour so existing setups keep working.
         self._routing = self._load_routing()
 
+        # Multi-model router (REQ-389). Loaded from .specsmith/local-models.yml
+        # if present; falls back to None (single-model mode) on any error.
+        self._model_router = self._load_model_router()
+
         # Consumers may swap this with a closure that routes through their
         # own bus (see ``serve._AgentThread``). The default writes JSONL.
         self._emit_event: Callable[..., None] = self._default_emit_event
@@ -383,15 +388,21 @@ class AgentRunner:
             for s in statuses:
                 if s.available:
                     lines.append(
-                        f"    {s.icon} {s.name:<10} ✓ ready   model: {s.model}  ({s.note})"
+                        f"    {s.icon} {s.name:<10} \u2713 ready   model: {s.model}  ({s.note})"
                     )
                 else:
-                    lines.append(f"    {s.icon} {s.name:<10} ✗ {s.note}")
+                    lines.append(f"    {s.icon} {s.name:<10} \u2717 {s.note}")
             lines.append("")
             if active_count == 0:
-                lines.append("  ⚠  No provider available — commands will return no response.")
+                lines.append("  \u26a0  No provider available — commands will return no response.")
             else:
-                lines.append("  Type plain English, or use /plan /ask /fix /test /commit /pr /exit")
+                # Show multi-model routing if the router is configured (REQ-389).
+                if self._model_router is not None:
+                    lines.append(self._model_router.table())
+                    lines.append("")
+                lines.append(
+                    "  Type plain English, or use /plan /ask /fix /test /commit /pr /models /exit"
+                )
             print("\n".join(lines), flush=True)
 
     def run_interactive(self) -> None:
@@ -447,6 +458,14 @@ class AgentRunner:
             self._history = []
             self._emit_event(type="system", message="History cleared.")
             return None
+        if text.strip() == "/models":
+            msg = (
+                self._model_router.table()
+                if self._model_router is not None
+                else "  (multi-model routing not configured — run: specsmith local-model setup)"
+            )
+            self._emit_event(type="system", message=msg)
+            return None
         if text.startswith("/model "):
             new_model = text.split(maxsplit=1)[1].strip()
             self.model = new_model
@@ -494,6 +513,26 @@ class AgentRunner:
             )
 
         block_id = self._next_block_id()
+
+        # ── Multi-model routing (REQ-389) ──────────────────────────────────
+        # Ask the router which Ollama model best fits this utterance, then
+        # temporarily set SPECSMITH_OLLAMA_MODEL so _run_ollama picks it up.
+        # The env var is restored (or removed) after the turn completes.
+        import os as _os
+
+        _prev_ollama_model = _os.environ.get("SPECSMITH_OLLAMA_MODEL")
+        if self._model_router is not None:
+            routed_model, switched = self._model_router.route(text)
+            if routed_model:
+                _os.environ["SPECSMITH_OLLAMA_MODEL"] = routed_model
+                if switched:
+                    role = self._model_router.active_role or "?"
+                    self._emit_event(
+                        type="system",
+                        message=f"\u21bb model \u2192 {routed_model} (role: {role})",
+                    )
+        # ──────────────────────────────────────────────────────────────────
+
         try:
             from specsmith.agent.chat_runner import (
                 DEFAULT_OLLAMA_HOST,
@@ -519,12 +558,16 @@ class AgentRunner:
                 recoverable=True,
             )
             return None
+        finally:
+            # Restore SPECSMITH_OLLAMA_MODEL to its pre-turn state (REQ-389).
+            if _prev_ollama_model is None:
+                _os.environ.pop("SPECSMITH_OLLAMA_MODEL", None)
+            else:
+                _os.environ["SPECSMITH_OLLAMA_MODEL"] = _prev_ollama_model
 
         if result is None:
             # All providers failed — give the user an actionable explanation
             # rather than silent emptiness.
-            import os as _os
-
             host = _os.environ.get("OLLAMA_HOST", DEFAULT_OLLAMA_HOST).rstrip("/")
             if _ollama_alive(host):
                 model = _pick_ollama_model(host)
@@ -655,6 +698,23 @@ class AgentRunner:
         except Exception:  # noqa: BLE001
             return None
 
+    def _load_model_router(self) -> Any | None:
+        """Load ModelRouter from .specsmith/local-models.yml (REQ-389).
+
+        Returns ``None`` when no multi-role config is found, preserving
+        single-model behaviour for existing setups.
+        """
+        try:
+            from specsmith.agent.model_router import ModelRouter
+            from specsmith.local_model import load_local_models_config
+
+            roles = load_local_models_config(self.project_dir)
+            if not roles:
+                return None
+            return ModelRouter(roles)
+        except Exception:  # noqa: BLE001
+            return None
+
     # ── Event plumbing ────────────────────────────────────────────────
 
     def _default_emit_event(self, **kwargs: Any) -> None: