nexus-toolkit 2.3.1__tar.gz → 2.3.2__tar.gz

{nexus_toolkit-2.3.1 → nexus_toolkit-2.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nexus-toolkit
-Version: 2.3.1
+Version: 2.3.2
 Summary: Hybrid search, retrieval, and Figma Make boilerplate generation MCP server
 Author-email: Ronald dela Cruz <rcdelacruz@users.noreply.github.com>
 License-Expression: MIT

{nexus_toolkit-2.3.1 → nexus_toolkit-2.3.2}/nexus_cli.py

@@ -99,7 +99,7 @@ _THEME = Theme({
 console = Console(theme=_THEME)
 
 
-_VERSION = "2.3.1"
+_VERSION = "2.3.2"
 
 _LOGO = """\
 [cyan]███╗   ██╗███████╗██╗  ██╗██╗   ██╗███████╗[/cyan]

{nexus_toolkit-2.3.1 → nexus_toolkit-2.3.2}/nexus_toolkit.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nexus-toolkit
-Version: 2.3.1
+Version: 2.3.2
 Summary: Hybrid search, retrieval, and Figma Make boilerplate generation MCP server
 Author-email: Ronald dela Cruz <rcdelacruz@users.noreply.github.com>
 License-Expression: MIT

{nexus_toolkit-2.3.1 → nexus_toolkit-2.3.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "nexus-toolkit"
-version = "2.3.1"
+version = "2.3.2"
 description = "Hybrid search, retrieval, and Figma Make boilerplate generation MCP server"
 authors = [
     {name = "Ronald dela Cruz", email = "rcdelacruz@users.noreply.github.com"}

{nexus_toolkit-2.3.1 → nexus_toolkit-2.3.2}/tools/devsecops/agent_runner.py

@@ -1,4 +1,3 @@
-import asyncio
 import json
 import logging
 import pathlib
@@ -114,13 +113,17 @@ def register_devsecops_tools(mcp: FastMCP) -> None:
         agent_name: str,
         context: str,
         workflow: str = "",
-        model: str = "claude-sonnet-4-6",
         memory_path: str = "",
         remember: bool = False,
     ) -> str:
         """
-        Run a named dev-workflow agent against provided context using the claude CLI.
-        Returns findings as JSON.
+        Load a dev-workflow agent and return its system prompt + formatted context
+        for the current Claude Code session to apply directly — no subprocess needed.
+
+        Claude Code MUST:
+        1. Adopt the returned system_prompt as its persona for this response
+        2. Respond fully to user_message as that agent (do not summarize or delegate)
+        3. If remember=True, call update_agent_memory(agent_name, memory_path, <findings>) after responding
 
         Args:
             agent_name: Name of the dev-workflow agent.
@@ -133,23 +136,16 @@ def register_devsecops_tools(mcp: FastMCP) -> None:
             context: Content to analyze (diff, file content, description, etc.)
             workflow: Optional workflow command name to prepend as instructions
                 (e.g. "workflow-review-code", "workflow-review-security")
-            model: Claude model to use (default: claude-sonnet-4-6)
             memory_path: Path to .claude/memory/agents/ directory in the target project.
                 When provided, past findings are injected as context before the run.
-            remember: When True, write findings back to memory after the run.
-                Requires memory_path. Writes to personal/ layer (team-safe).
+            remember: When True, Claude Code must call update_agent_memory after responding
+                to persist findings. Requires memory_path.
         """
         try:
             system_prompt = _load_dev_agent(agent_name)
         except FileNotFoundError as e:
             return json.dumps({"error": str(e)})
 
-        claude_bin = _find_claude()
-        if not claude_bin:
-            return json.dumps({
-                "error": "claude CLI not found. Install Claude Code or ensure it is on PATH."
-            })
-
         # Build user message — optionally prepend workflow instructions
         user_message = context
         if workflow:
@@ -163,86 +159,32 @@ def register_devsecops_tools(mcp: FastMCP) -> None:
                 )
 
         # Prepend past memory if memory_path is provided
+        memory_context = ""
         if memory_path:
             memory_block = build_memory_context_block(memory_path, agent_name)
             if memory_block:
+                memory_context = memory_block
                 user_message = memory_block + user_message
 
-        cmd = [
-            claude_bin,
-            "--print",
-            "--dangerously-skip-permissions",
-            "--model", model,
-            "--system-prompt", system_prompt,
-            user_message,
-        ]
-
-        # Idle timeout: if the process produces no output for this many seconds
-        # it is considered stuck. Active generation keeps resetting the timer,
-        # so a lengthy-but-healthy run is never killed prematurely.
-        idle_timeout = 240  # seconds — generous for agents like product-manager that think before writing
-
-        try:
-            proc = await asyncio.create_subprocess_exec(
-                *cmd,
-                stdout=asyncio.subprocess.PIPE,
-                stderr=asyncio.subprocess.PIPE,
+        after_response = []
+        if remember and memory_path:
+            after_response.append(
+                f"Call update_agent_memory(agent_name='{agent_name}', "
+                f"memory_path='{memory_path}', content=<your findings>) to persist findings."
             )
 
-            chunks: list[str] = []
-            assert proc.stdout is not None
-            while True:
-                try:
-                    chunk = await asyncio.wait_for(proc.stdout.read(4096), timeout=idle_timeout)
-                except asyncio.TimeoutError:
-                    proc.kill()
-                    await proc.wait()
-                    logger.error(
-                        "run_agent: claude CLI idle for %ds with no output (agent=%s)",
-                        idle_timeout, agent_name,
-                    )
-                    return json.dumps({
-                        "error": (
-                            f"The run_agent call timed out after {idle_timeout}s of inactivity. "
-                            "The agent uses the claude CLI under the hood to run, and it produced "
-                            "no output during that window. This could be due to the agent taking "
-                            "too long to respond, or an issue with the underlying claude CLI invocation."
-                        )
-                    })
-                if not chunk:
-                    break  # EOF — process finished
-                chunks.append(chunk.decode())
-
-            await proc.wait()
-            stderr_text = (await proc.stderr.read()).decode().strip()
-
-            if proc.returncode != 0:
-                logger.error("claude CLI error for agent '%s': %s", agent_name, stderr_text)
-                return json.dumps({
-                    "error": f"claude exited with code {proc.returncode}",
-                    "detail": stderr_text,
-                })
-
-            findings = "".join(chunks).strip()
-
-            # Write findings back to memory if requested
-            pending_review: list[str] = []
-            memory_updated = False
-            if memory_path and remember:
-                memory_updated, pending_review = update_memory(memory_path, agent_name, findings)
-
-            return json.dumps({
-                "agent": agent_name,
-                "workflow": workflow or None,
-                "findings": findings,
-                "model": model,
-                "memory_updated": memory_updated,
-                "pending_review": pending_review,
-            })
-
-        except Exception as exc:
-            logger.error("run_agent error for '%s': %s", agent_name, exc)
-            return json.dumps({"error": str(exc)})
+        return json.dumps({
+            "agent": agent_name,
+            "workflow": workflow or None,
+            "system_prompt": system_prompt,
+            "user_message": user_message,
+            "has_memory_context": bool(memory_context),
+            "after_response": after_response,
+            "_instruction": (
+                f"Adopt system_prompt as your persona and respond fully to user_message as the {agent_name} agent. "
+                + (" ".join(after_response) if after_response else "")
+            ),
+        })
 
     @mcp.tool()
     async def get_agent_memory(