crucible-mcp 1.0.1__tar.gz → 1.2.0__tar.gz

{crucible_mcp-1.0.1 → crucible_mcp-1.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: crucible-mcp
-Version: 1.0.1
+Version: 1.2.0
 Summary: Code review MCP server for Claude. Not affiliated with Atlassian.
 Author: be.nvy
 License-Expression: MIT
@@ -58,6 +58,20 @@ That's it. Crucible will now:
 2. Review files Claude edits (Claude Code hook)
 3. Block code that violates bundled assertions (security, error handling, smart contracts)
 
+## CLAUDE.md Setup
+
+Add to your `CLAUDE.md` to inject rules at session start:
+
+```markdown
+# Project
+
+At session start, call get_assertions() to load enforcement rules.
+
+For code review: crucible review
+```
+
+This tells Claude to load the active assertions *before* writing code, not just catch violations after.
+
 ## How Enforcement Works
 
 ```
@@ -109,13 +123,16 @@ Add to Claude Code (`.mcp.json`):
 
 | Tool | Purpose |
 |------|---------|
+| `get_assertions()` | **Session start:** Load enforced patterns into context |
+| `get_principles(topic)` | **Session start:** Load engineering knowledge by topic |
+| `load_knowledge(files)` | **Session start:** Load specific knowledge files |
 | `review(path)` | Full review: analysis + skills + knowledge + assertions |
 | `review(mode='staged')` | Review git changes with enforcement |
-| `load_knowledge(files)` | Load specific knowledge files |
-| `get_principles(topic)` | Load engineering knowledge by topic |
 | `delegate_*` | Direct tool access (semgrep, ruff, slither, bandit) |
 | `check_tools()` | Show installed analysis tools |
 
+**Tip:** Call `get_assertions()` at the start of a session so Claude knows what patterns to avoid *before* writing code.
+
 ## CLI
 
 ```bash

{crucible_mcp-1.0.1 → crucible_mcp-1.2.0}/README.md

@@ -40,6 +40,20 @@ That's it. Crucible will now:
 2. Review files Claude edits (Claude Code hook)
 3. Block code that violates bundled assertions (security, error handling, smart contracts)
 
+## CLAUDE.md Setup
+
+Add to your `CLAUDE.md` to inject rules at session start:
+
+```markdown
+# Project
+
+At session start, call get_assertions() to load enforcement rules.
+
+For code review: crucible review
+```
+
+This tells Claude to load the active assertions *before* writing code, not just catch violations after.
+
 ## How Enforcement Works
 
 ```
@@ -91,13 +105,16 @@ Add to Claude Code (`.mcp.json`):
 
 | Tool | Purpose |
 |------|---------|
+| `get_assertions()` | **Session start:** Load enforced patterns into context |
+| `get_principles(topic)` | **Session start:** Load engineering knowledge by topic |
+| `load_knowledge(files)` | **Session start:** Load specific knowledge files |
 | `review(path)` | Full review: analysis + skills + knowledge + assertions |
 | `review(mode='staged')` | Review git changes with enforcement |
-| `load_knowledge(files)` | Load specific knowledge files |
-| `get_principles(topic)` | Load engineering knowledge by topic |
 | `delegate_*` | Direct tool access (semgrep, ruff, slither, bandit) |
 | `check_tools()` | Show installed analysis tools |
 
+**Tip:** Call `get_assertions()` at the start of a session so Claude knows what patterns to avoid *before* writing code.
+
 ## CLI
 
 ```bash

{crucible_mcp-1.0.1 → crucible_mcp-1.2.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "crucible-mcp"
-version = "1.0.1"
+version = "1.2.0"
 description = "Code review MCP server for Claude. Not affiliated with Atlassian."
 readme = "README.md"
 requires-python = ">=3.11"

{crucible_mcp-1.0.1 → crucible_mcp-1.2.0}/src/crucible/knowledge/loader.py

@@ -247,14 +247,17 @@ def get_custom_knowledge_files() -> set[str]:
 
 
 def load_all_knowledge(
-    include_bundled: bool = False,
+    include_bundled: bool = True,
     filenames: set[str] | None = None,
 ) -> tuple[list[str], str]:
     """Load multiple knowledge files.
 
+    Knowledge follows cascade priority: project > user > bundled.
+    Project/user files override bundled files with the same name.
+
     Args:
-        include_bundled: If True, include bundled knowledge files
-        filenames: Specific files to load (if None, loads based on include_bundled)
+        include_bundled: If True, include bundled knowledge files (default: True)
+        filenames: Specific files to load (if None, loads all from cascade)
 
     Returns:
         Tuple of (list of loaded filenames, combined content)

{crucible_mcp-1.0.1 → crucible_mcp-1.2.0}/src/crucible/models.py

@@ -61,16 +61,3 @@ DOMAIN_HEURISTICS: dict[Domain, dict[str, list[str]]] = {
 }
 
 
-@dataclass(frozen=True)
-class FullReviewResult:
-    """Result from full_review tool."""
-
-    domains_detected: tuple[str, ...]
-    severity_summary: dict[str, int]
-    findings: tuple[ToolFinding, ...]
-    applicable_skills: tuple[str, ...]
-    skill_triggers_matched: dict[str, tuple[str, ...]]
-    principles_loaded: tuple[str, ...]
-    principles_content: str
-    sage_knowledge: str | None = None
-    sage_query_used: str | None = None