npcsh 1.1.16__py3-none-any.whl → 1.1.18__py3-none-any.whl

npcsh-1.1.18.data/data/npcsh/npc_team/benchmark.jinx

@@ -0,0 +1,146 @@
+jinx_name: benchmark
+description: Run Terminal-Bench evaluation to benchmark npcsh performance with different models
+inputs:
+- model: ""
+- provider: ""
+- action: "check"
+- concurrent: "4"
+- npc_name: ""
+
+steps:
+  - name: run_benchmark
+    engine: python
+    code: |
+      import os
+      import sys
+
+      action = {{ action | default("check") | tojson }}.strip().lower()
+      model = {{ model | default("") | tojson }}.strip()
+      provider = {{ provider | default("") | tojson }}.strip()
+      concurrent = int({{ concurrent | default("4") | tojson }} or "4")
+      npc_name_input = {{ npc_name | default("") | tojson }}.strip() or None
+
+      if not model:
+          model = npc.model if npc and npc.model else ""
+      if not provider:
+          provider = npc.provider if npc and npc.provider else "anthropic"
+
+      try:
+          from npcsh.benchmark import BenchmarkRunner, run_benchmark
+          BENCHMARK_AVAILABLE = True
+      except ImportError:
+          BENCHMARK_AVAILABLE = False
+
+      if action == "check":
+          output = "## Terminal-Bench Integration Status\n\n"
+
+          if not BENCHMARK_AVAILABLE:
+              output += "**Status:** Benchmark module not fully loaded (harbor not installed)\n\n"
+          else:
+              output += "**Status:** Ready\n\n"
+
+          if BENCHMARK_AVAILABLE:
+              runner = BenchmarkRunner()
+              deps = runner.check_dependencies()
+              output += "### Dependencies:\n"
+              for dep, installed in deps.items():
+                  status = "Installed" if installed else "Not installed"
+                  output += "- **{}**: {}\n".format(dep, status)
+
+              if not all(deps.values()):
+                  output += "\n### Installation:\n"
+                  output += "```bash\n"
+                  output += "pip install harbor terminal-bench\n"
+                  output += "```\n"
+
+          output += "\n### Usage:\n"
+          output += "```\n"
+          output += "/benchmark action=quick\n"
+          output += "/benchmark action=run model=gpt-4o provider=openai\n"
+          output += "/benchmark action=list\n"
+          output += "```\n"
+
+      elif action == "list":
+          if not BENCHMARK_AVAILABLE:
+              output = "Error: Benchmark module not available. Run `/benchmark` first."
+          else:
+              runner = BenchmarkRunner()
+              runs = runner.list_past_runs()
+
+              if not runs:
+                  output = "No past benchmark runs found."
+              else:
+                  output = "## Past Benchmark Runs ({} total)\n\n".format(len(runs))
+                  for run in runs[:10]:
+                      timestamp = run.get('timestamp', 'unknown')[:19]
+                      model_name = run.get('model', 'unknown')
+                      result = run.get('result', {})
+                      accuracy = result.get('accuracy', 0)
+                      passed = result.get('passed_tasks', 0)
+                      total = result.get('total_tasks', 0)
+
+                      output += "### {}\n".format(timestamp)
+                      output += "- **Model:** {}\n".format(model_name)
+                      output += "- **Accuracy:** {:.1%}\n".format(accuracy)
+                      output += "- **Tasks:** {}/{}\n\n".format(passed, total)
+
+      elif action == "quick":
+          if not BENCHMARK_AVAILABLE:
+              output = "Error: Install with: pip install harbor terminal-bench"
+          else:
+              output = "## Quick Test: {}/{}\n\n".format(provider, model)
+              output += "Running quick test with 3 tasks...\n\n"
+
+              try:
+                  from npcsh.benchmark import quick_test
+                  result = quick_test(model=model, provider=provider)
+
+                  status = "PASS" if result.success else "FAIL"
+                  output += "**Status:** {}\n".format(status)
+                  output += "**Accuracy:** {:.1%}\n".format(result.accuracy)
+                  output += "**Tasks:** {}/{}\n".format(result.passed_tasks, result.total_tasks)
+                  output += "**Duration:** {:.1f}s\n".format(result.duration_seconds)
+
+                  if result.error:
+                      output += "\n**Error:** {}\n".format(result.error)
+
+                  output += "\n**Output:** {}\n".format(result.output_dir)
+
+              except Exception as e:
+                  output = "Error running quick test: {}".format(e)
+
+      elif action == "run":
+          if not BENCHMARK_AVAILABLE:
+              output = "Error: Install with: pip install harbor terminal-bench"
+          else:
+              output = "## Benchmark Run: {}/{}\n\n".format(provider, model)
+              output += "Running Terminal-Bench 2.0 with {} concurrent tasks...\n\n".format(concurrent)
+
+              try:
+                  runner = BenchmarkRunner()
+                  result = runner.run(
+                      model=model,
+                      provider=provider,
+                      n_concurrent=concurrent,
+                      npc_name=npc_name_input,
+                  )
+
+                  status = "SUCCESS" if result.success else "FAILED"
+                  output += "**Status:** {}\n".format(status)
+                  output += "**Accuracy:** {:.1%}\n".format(result.accuracy)
+                  output += "**Tasks Passed:** {}/{}\n".format(result.passed_tasks, result.total_tasks)
+                  output += "**Duration:** {:.1f}s\n".format(result.duration_seconds)
+                  output += "**Total Tokens:** {:,}\n".format(result.total_tokens)
+                  output += "**Total Cost:** ${:.4f}\n".format(result.total_cost_usd)
+
+                  if result.error:
+                      output += "\n**Error:** {}\n".format(result.error)
+
+                  output += "\n**Results saved to:** {}\n".format(result.output_dir)
+
+              except Exception as e:
+                  import traceback
+                  output = "Error running benchmark: {}\n\n{}".format(e, traceback.format_exc())
+
+      else:
+          output = "Unknown action: {}\n\nAvailable: check, run, quick, list".format(action)

{npcsh-1.1.16.data → npcsh-1.1.18.data}/data/npcsh/npc_team/browser_action.jinx

@@ -12,14 +12,14 @@ description: |
   - get_page: Get page title, URL, and visible text
   - get_elements: Get interactive elements with their selectors
   - press_key: Press a key (enter, tab, escape, etc)
-  Selectors: CSS (#id, .class, input[name="x"]) or xpath://... for XPath
+  Selectors: CSS (
 inputs:
-  - action:
+- action:
       description: "Action: click, type, type_and_enter, set_value, select, wait, scroll, get_text, get_page, get_elements, press_key"
-  - selector:
+- selector:
       description: "CSS selector or XPath (prefix xpath: for XPath)"
       default: ""
-  - value:
+- value:
       description: "Value for type/select, or scroll direction, or key name"
       default: ""
 

{npcsh-1.1.16.data → npcsh-1.1.18.data}/data/npcsh/npc_team/browser_screenshot.jinx

@@ -1,7 +1,7 @@
 jinx_name: browser_screenshot
 description: Take a screenshot of the current browser page.
 inputs:
-  - filename:
+- filename:
       description: "Optional filename for screenshot"
       default: ""
 

{npcsh-1.1.16.data → npcsh-1.1.18.data}/data/npcsh/npc_team/build.jinx

@@ -1,11 +1,11 @@
 jinx_name: "build"
 description: "Build deployment artifacts for NPC team"
 inputs:
-  - target: "flask" # The type of deployment target (e.g., flask, docker, cli, static).
-  - outdir: "./build" # The output directory for built artifacts.
-  - team: "./npc_team" # The path to the NPC team directory.
-  - port: 5337 # The port for flask server builds.
-  - cors: "" # Comma-separated CORS origins for flask server builds.
+- target: "flask"
+- outdir: "./build"
+- team: "./npc_team"
+- port: 5337
+- cors: ""
 steps:
   - name: "execute_build"
     engine: "python"

{npcsh-1.1.16.data → npcsh-1.1.18.data}/data/npcsh/npc_team/chat.jinx

@@ -1,10 +1,10 @@
 jinx_name: chat
 description: Simple chat mode - LLM conversation without tool execution
 inputs:
-  - query: null
-  - model: null
-  - provider: null
-  - stream: true
+- query: null
+- model: null
+- provider: null
+- stream: true
 
 steps:
   - name: chat_response

{npcsh-1.1.16.data → npcsh-1.1.18.data}/data/npcsh/npc_team/click.jinx

@@ -1,8 +1,8 @@
 jinx_name: click
 description: Click at screen coordinates (0-100 percentage)
 inputs:
-  - x: 50  # X coordinate as percentage (0-100)
-  - y: 50  # Y coordinate as percentage (0-100)
+- x: 50
+- y: 50
 
 steps:
   - name: perform_click

npcsh-1.1.18.data/data/npcsh/npc_team/close_pane.jinx

@@ -0,0 +1,9 @@
+jinx_name: studio.close_pane
+description: Close a pane in NPC Studio. Use paneId="active" or omit to close the active pane.
+inputs:
+- paneId: "active"
+steps:
+  - name: frontend_action
+    engine: python
+    code: |
+      context['output'] = "Action executed by frontend"

npcsh-1.1.18.data/data/npcsh/npc_team/close_tab.jinx

@@ -0,0 +1,10 @@
+jinx_name: studio.close_tab
+description: Close a specific tab in a pane.
+inputs:
+- paneId: "active"
+- tabIndex: 0
+steps:
+  - name: frontend_action
+    engine: python
+    code: |
+      context['output'] = "Action executed by frontend"

{npcsh-1.1.16.data → npcsh-1.1.18.data}/data/npcsh/npc_team/cmd.jinx

@@ -1,10 +1,10 @@
 jinx_name: cmd
 description: Command mode - LLM generates and executes shell commands
 inputs:
-  - query: null
-  - model: null
-  - provider: null
-  - stream: true
+- query: null
+- model: null
+- provider: null
+- stream: true
 
 steps:
   - name: cmd_execute

{npcsh-1.1.16.data → npcsh-1.1.18.data}/data/npcsh/npc_team/compile.jinx

@@ -1,8 +1,8 @@
 jinx_name: "compile"
 description: "Compile NPC profiles"
 inputs:
-  - npc_file_path: "" # Optional path to a specific NPC file to compile.
-  - npc_team_dir: "./npc_team" # Directory containing NPC profiles to compile, if no specific file is given.
+- npc_file_path: ""
+- npc_team_dir: "./npc_team"
 steps:
   - name: "compile_npcs"
     engine: "python"

{npcsh-1.1.16.data → npcsh-1.1.18.data}/data/npcsh/npc_team/compress.jinx

@@ -1,12 +1,12 @@
 jinx_name: "compress"
 description: "Manages conversation and knowledge context. Defaults to compacting context. Use flags for other operations."
 inputs:
-  - flush: ""       # The number of recent messages to flush.
-  - sleep: False    # If true, evolves the knowledge graph.
-  - dream: False    # Used with --sleep. Runs creative synthesis.
-  - ops: ""         # Used with --sleep. Comma-separated list of KG operations.
-  - model: ""       # Used with --sleep. LLM model for KG evolution.
-  - provider: ""    # Used with --sleep. LLM provider for KG evolution.
+- flush: ""
+- sleep: False
+- dream: False
+- ops: ""
+- model: ""
+- provider: ""
 steps:
   - name: "manage_context_and_memory"
     engine: "python"
@@ -53,8 +53,8 @@ steps:
           operations_config = [op.strip() for op in operations_str.split(',')] if operations_str else None
           if not llm_model and current_npc: llm_model = current_npc.model
           if not llm_provider and current_npc: llm_provider = current_npc.provider
-          if not llm_model: llm_model = "gemini-1.5-pro"
-          if not llm_provider: llm_provider = "gemini"
+          if not llm_model: llm_model = state.chat_model if state else "llama3.2"
+          if not llm_provider: llm_provider = state.chat_provider if state else "ollama"
 
           team_name = current_team.name if current_team else "__none__"
           npc_name = current_npc.name if current_npc else "__none__"

npcsh-1.1.18.data/data/npcsh/npc_team/confirm.jinx

@@ -0,0 +1,10 @@
+jinx_name: studio.confirm
+description: Show a confirmation dialog and return the user's choice.
+inputs:
+- message: ""
+- title: "Confirm"
+steps:
+  - name: frontend_action
+    engine: python
+    code: |
+      context['output'] = "Action executed by frontend"

{npcsh-1.1.16.data → npcsh-1.1.18.data}/data/npcsh/npc_team/convene.jinx

@@ -1,11 +1,11 @@
 jinx_name: convene
 description: Run a cycle of discussions between NPCs on a topic. The orchestrator convenes agents to discuss and synthesize.
 inputs:
-  - topic: ""
-  - npcs: "alicanto,corca,guac"
-  - rounds: 3
-  - model: null
-  - provider: null
+- topic: ""
+- npcs: "alicanto,corca,guac"
+- rounds: 3
+- model: null
+- provider: null
 steps:
   - name: convene_discussion
     engine: python
@@ -21,8 +21,8 @@ steps:
       team = context.get('team')
       messages = context.get('messages', [])
 
-      model = context.get('model') or (npc.model if npc else 'gemini-1.5-flash')
-      provider = context.get('provider') or (npc.provider if npc else 'gemini')
+      model = context.get('model') or (npc.model if npc else (state.chat_model if state else 'llama3.2'))
+      provider = context.get('provider') or (npc.provider if npc else (state.chat_provider if state else 'ollama'))
 
       if not topic:
           context['output'] = """Usage: /convene <topic>