specsmith 0.4.0.dev222__tar.gz → 0.4.0.dev223__tar.gz

{specsmith-0.4.0.dev222/src/specsmith.egg-info → specsmith-0.4.0.dev223}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: specsmith
-Version: 0.4.0.dev222
+Version: 0.4.0.dev223
 Summary: Applied Epistemic Engineering toolkit — AEE agent sessions, execution profiles, FPGA/HDL governance, tool installer, 50+ CLI commands.
 Author: BitConcepts
 License-Expression: MIT

{specsmith-0.4.0.dev222 → specsmith-0.4.0.dev223}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "specsmith"
-version = "0.4.0.dev222"
+version = "0.4.0.dev223"
 description = "Applied Epistemic Engineering toolkit — AEE agent sessions, execution profiles, FPGA/HDL governance, tool installer, 50+ CLI commands."
 readme = "README.md"
 license = "MIT"
@@ -107,6 +107,11 @@ select = ["E", "F", "W", "I", "UP", "B", "SIM"]
 # Long rule text and install command strings in documentation modules
 "src/specsmith/toolrules.py" = ["E501"]
 "src/specsmith/tool_installer.py" = ["E501"]
+# One-shot rebuild scripts contain literal REQ description strings that are
+# intentionally long; wrapping them would damage readability of the source
+# of truth they reconstruct.
+"scripts/rebuild_requirements_json.py" = ["E501"]
+"scripts/rebuild_requirements_md.py" = ["E501"]
 
 [tool.mypy]
 python_version = "3.10"
@@ -141,9 +146,11 @@ ignore_missing_imports = true
 module = ["epistemic", "epistemic.*"]
 ignore_errors = true
 
-# New modules that use dynamic typing patterns incompatible with strict mypy.
+# Modules that use dynamic typing patterns incompatible with strict mypy.
 # These are correct at runtime; the Any-heavy urllib, json.loads, etc. patterns
 # are the source of most errors here. Exclude from strict type checking.
+# REQ-111: shrunk this list as part of the pre-1.0 cleanup; broker, safety,
+# console_utils, and indexer have been graduated to strict mypy.
 [[tool.mypy.overrides]]
 module = [
     "specsmith.ollama_cmds",
@@ -153,14 +160,10 @@ module = [
     "specsmith.importer",
     "specsmith.agent.providers.gemini",
     "specsmith.agent.runner",
-    "specsmith.agent.broker",
     "specsmith.agent.cleanup",
-    "specsmith.agent.indexer",
     "specsmith.agent.orchestrator",
     "specsmith.agent.repl",
-    "specsmith.agent.safety",
     "specsmith.agent.tools",
-    "specsmith.console_utils",
     "specsmith.profiles",
     "specsmith.serve",
     "specsmith.toolrules",

{specsmith-0.4.0.dev222 → specsmith-0.4.0.dev223}/src/specsmith/agent/broker.py

@@ -40,6 +40,7 @@ from collections.abc import Callable
 from dataclasses import dataclass, field
 from enum import Enum
 from pathlib import Path
+from typing import Any
 
 # ---------------------------------------------------------------------------
 # Intent classification
@@ -285,7 +286,7 @@ def infer_scope(
 class PreflightDecision:
     """Wrapped Specsmith preflight outcome."""
 
-    raw: dict
+    raw: dict[str, Any]
     decision: str = "unknown"
     work_item_id: str = ""
     requirement_ids: list[str] = field(default_factory=list)
@@ -294,7 +295,7 @@ class PreflightDecision:
     instruction: str = ""
 
     @classmethod
-    def from_json(cls, payload: dict) -> PreflightDecision:
+    def from_json(cls, payload: dict[str, Any]) -> PreflightDecision:
         return cls(
             raw=payload,
             decision=str(payload.get("decision", "unknown")),
@@ -476,7 +477,7 @@ RETRY_STRATEGIES = (
 )
 
 
-def classify_retry_strategy(report: dict, decision: PreflightDecision) -> str:
+def classify_retry_strategy(report: dict[str, Any], decision: PreflightDecision) -> str:
     """Map an executor failure report to one of the canonical retry strategies.
 
     The classification is deterministic and inspects:
@@ -529,7 +530,7 @@ def classify_retry_strategy(report: dict, decision: PreflightDecision) -> str:
 def execute_with_governance(
     decision: PreflightDecision,
     *,
-    executor: Callable[[PreflightDecision, int], dict],
+    executor: Callable[[PreflightDecision, int], dict[str, Any]],
     retry_budget: int = DEFAULT_RETRY_BUDGET,
 ) -> RunResult:
     """Run the work with Specsmith governance and a hard retry budget.
@@ -546,7 +547,7 @@ def execute_with_governance(
 
     last_summary = ""
     last_confidence = 0.0
-    last_report: dict = {}
+    last_report: dict[str, Any] = {}
     for attempt in range(1, retry_budget + 1):
         report = executor(decision, attempt) or {}
         last_report = report

{specsmith-0.4.0.dev222 → specsmith-0.4.0.dev223}/src/specsmith/agent/indexer.py

@@ -5,7 +5,7 @@ import subprocess
 from pathlib import Path
 
 
-def generate_index(cwd: str = None):
+def generate_index(cwd: str | None = None) -> None:
     """Generate repository index into .repo-index/"""
     if cwd is None:
         cwd = os.getcwd()

{specsmith-0.4.0.dev222 → specsmith-0.4.0.dev223}/src/specsmith/agent/orchestrator.py

@@ -207,13 +207,15 @@ Next action:
 
         AG2's ``initiate_chat`` returns a ``ChatResult`` whose ``summary`` is
         the last assistant message and whose ``chat_history`` lists every
-        turn. We treat reaching a non-empty summary that includes the
-        ``Next action:`` section as equilibrium per REQ-073, and parse
-        ``Files changed:`` and ``Test results:`` if present. The confidence
-        is conservative (0.85 for a complete contract response, 0.4 for a
-        partial one) and is meant to be replaced by a real verifier signal
-        once one is wired in.
+        turn. We parse the Nexus output contract out of the summary and feed
+        the structured signal through :func:`specsmith.agent.verifier.score`
+        (REQ-108) so ``equilibrium`` and ``confidence`` reflect real test /
+        ruff / mypy state instead of a hardcoded heuristic. When the LLM
+        returns no contract sections at all we fall back to the previous
+        conservative defaults so behaviour stays the same on degraded runs.
         """
+        from specsmith.agent.verifier import report_from_chat_sections, score
+
         summary = ""
         if chat_result is not None:
             summary = getattr(chat_result, "summary", "") or ""
@@ -221,8 +223,6 @@ Next action:
                 summary = str(summary)
 
         sections = self._parse_output_contract(summary)
-        equilibrium = bool(sections) and "next_action" in sections
-        confidence = 0.85 if equilibrium else (0.4 if summary else 0.0)
 
         files_changed: list[str] = []
         files_section = sections.get("files_changed", "")
@@ -236,6 +236,22 @@ Next action:
         if tests_section:
             test_results["raw"] = tests_section
 
+        # REQ-108: derive confidence and equilibrium from the real verifier
+        # signal rather than guessing from the presence of contract sections.
+        if sections:
+            report = report_from_chat_sections(sections, files_changed=files_changed)
+            verdict = score(report, confidence_target=0.7)
+            confidence = verdict.confidence
+            # Treat "reached the next_action section" as a soft floor for
+            # equilibrium so the harness still distinguishes a structurally
+            # complete contract from a totally empty one.
+            equilibrium = verdict.equilibrium or (
+                "next_action" in sections and report.has_changes and report.test_failed == 0
+            )
+        else:
+            equilibrium = False
+            confidence = 0.4 if summary else 0.0
+
         return TaskResult(
             equilibrium=equilibrium,
             confidence=confidence,

{specsmith-0.4.0.dev222 → specsmith-0.4.0.dev223}/src/specsmith/console_utils.py

@@ -23,11 +23,12 @@ from __future__ import annotations
 
 import os
 import sys
+from typing import IO, Any
 
 from rich.console import Console
 
 
-def _ensure_utf8_stream(stream) -> None:
+def _ensure_utf8_stream(stream: IO[str]) -> None:
     """Best-effort upgrade of ``stream`` to UTF-8 with safe error handling."""
     reconfigure = getattr(stream, "reconfigure", None)
     if reconfigure is None:
@@ -40,7 +41,7 @@ def _ensure_utf8_stream(stream) -> None:
         return
 
 
-def make_console(**kwargs) -> Console:
+def make_console(**kwargs: Any) -> Console:
     """Return a UTF-8 safe ``rich.console.Console``.
 
     Any keyword arguments are forwarded to ``Console`` and override the
@@ -53,7 +54,7 @@ def make_console(**kwargs) -> Console:
     # we may spawn (e.g. tests, CI runners).
     os.environ.setdefault("PYTHONIOENCODING", "utf-8")
 
-    defaults = {
+    defaults: dict[str, Any] = {
         "legacy_windows": False,
         "soft_wrap": False,
         "force_terminal": None,

{specsmith-0.4.0.dev222 → specsmith-0.4.0.dev223/src/specsmith.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: specsmith
-Version: 0.4.0.dev222
+Version: 0.4.0.dev223
 Summary: Applied Epistemic Engineering toolkit — AEE agent sessions, execution profiles, FPGA/HDL governance, tool installer, 50+ CLI commands.
 Author: BitConcepts
 License-Expression: MIT