diff --git a/.gitignore b/.gitignore index 3bedddb4..bbdfeea3 100644 --- a/.gitignore +++ b/.gitignore @@ -107,3 +107,8 @@ credentials.json # LangGraph API server .langgraph_api/ + +# Reports (generated artifacts only — anchored to repo root) +/report*.md +/report*.json +/report*.sarif diff --git a/README.md b/README.md index 4a09b50b..c5c3f5ae 100644 --- a/README.md +++ b/README.md @@ -226,6 +226,12 @@ export SKILLSPECTOR_PROVIDER=nv_build export NVIDIA_INFERENCE_KEY=nvapi-... skillspector scan ./my-skill/ +# Gemini (via OpenAI compatibility layer) +export SKILLSPECTOR_PROVIDER=openai +export OPENAI_API_KEY="YOUR_GEMINI_API_KEY" +export OPENAI_BASE_URL="https://generativelanguage.googleapis.com/v1beta/openai/" +export SKILLSPECTOR_MODEL=gemini-3.5-flash + # Local Claude CLI — no API key; uses your existing `claude auth login` session # Requires: claude CLI installed and authenticated (claude auth login) export SKILLSPECTOR_PROVIDER=claude_cli diff --git a/model_registry.yaml b/model_registry.yaml index e1c2b8c6..7a1e34f8 100644 --- a/model_registry.yaml +++ b/model_registry.yaml @@ -40,3 +40,7 @@ models: "openai/openai/gpt-5.3-chat": context_length: 128000 max_output_tokens: 16384 + + "gemini-3.5-flash": + context_length: 1048576 + max_output_tokens: 8192 diff --git a/src/skillspector/cli.py b/src/skillspector/cli.py index 9b9a9b5e..e9f7754f 100644 --- a/src/skillspector/cli.py +++ b/src/skillspector/cli.py @@ -25,6 +25,7 @@ import os import shutil import sys +import warnings from enum import StrEnum from pathlib import Path from typing import Annotated @@ -32,11 +33,14 @@ import typer from langchain_core.runnables import RunnableConfig from rich.console import Console +from rich.progress import BarColumn, Progress, SpinnerColumn, TextColumn, TimeElapsedColumn +from rich.tree import Tree from skillspector import __version__ from skillspector.graph import graph from skillspector.logging_config import get_logger, set_level from skillspector.multi_skill import MultiSkillDetectionResult, detect_skills +from skillspector.nodes.analyzers import ANALYZER_NODE_IDS from skillspector.suppression import build_baseline_dict, dump_baseline, load_baseline logger = get_logger(__name__) @@ -318,7 +322,80 @@ def scan( not no_llm, ) trace_config = _build_trace_config(input_path, format, no_llm) - result = graph.invoke(state, config=trace_config) + if verbose: + result = graph.invoke(state, config=trace_config) + else: + total_steps = 4 + len(ANALYZER_NODE_IDS) + result = dict(state) + + # Use stderr for progress so stdout remains clean for structured outputs + err_console = Console(stderr=True) + + # Suppress noisy Pydantic serialization warnings scoped to the graph run + with warnings.catch_warnings(), Progress( + SpinnerColumn(), + TextColumn("[progress.description]{task.description}"), + BarColumn(), + TextColumn("[progress.percentage]{task.percentage:>3.0f}%"), + TimeElapsedColumn(), + console=err_console, + transient=True, + ) as progress: + warnings.filterwarnings("ignore", category=UserWarning, module="pydantic") + task_id = progress.add_task("Resolving input...", total=total_steps) + + num_files = 0 + analyzers_done = 0 + total_analyzers = len(ANALYZER_NODE_IDS) + + for update in graph.stream(state, config=trace_config, stream_mode="updates"): + for node_name, node_output in update.items(): + progress.advance(task_id) + + # Accumulate scalar outputs needed by the CLI (report_body, risk_score, temp_dir, sarif_report) + if "temp_dir_for_cleanup" in node_output: + result["temp_dir_for_cleanup"] = node_output["temp_dir_for_cleanup"] + if "report_body" in node_output: + result["report_body"] = node_output["report_body"] + if "sarif_report" in node_output: + result["sarif_report"] = node_output["sarif_report"] + if "risk_score" in node_output: + result["risk_score"] = node_output["risk_score"] + + # Update UI text based on graph progression + if node_name == "resolve_input": + progress.update(task_id, description="Building context...") + elif node_name == "build_context": + components = node_output.get("components", []) + num_files = len(components) + progress.update(task_id, description=f"Analyzing {num_files} files (0/{total_analyzers} rules applied)...") + + # Print a proper report of the files and directories being scanned + tree = Tree("[bold blue]Discovered Files to Scan[/bold blue]") + nodes = {"": tree} + for path in sorted(components): + parts = Path(path).parts + current = "" + for part in parts: + parent = current + current = f"{current}/{part}" if current else part + if current not in nodes: + is_file = current == path + icon = "📄 " if is_file else "📁 " + style = "green" if is_file else "cyan" + nodes[current] = nodes[parent].add(f"[{style}]{icon}{part}[/{style}]") + + err_console.print(tree) + err_console.print() + + elif node_name in ANALYZER_NODE_IDS: + analyzers_done += 1 + progress.update(task_id, description=f"Analyzing {num_files} files ({analyzers_done}/{total_analyzers} rules applied)...") + # Print which rule just finished above the progress bar + err_console.print(f"[dim]✔ Rule completed: {node_name}[/dim]") + elif node_name == "meta_analyzer": + progress.update(task_id, description="Generating report...") + err_console.print("[dim]✔ Rule completed: meta_analyzer (filtering findings)[/dim]") _write_result(result, output, format) diff --git a/src/skillspector/providers/openai/model_registry.yaml b/src/skillspector/providers/openai/model_registry.yaml index a4d26067..a539cccd 100644 --- a/src/skillspector/providers/openai/model_registry.yaml +++ b/src/skillspector/providers/openai/model_registry.yaml @@ -12,3 +12,7 @@ models: "gpt-5.4": context_length: 1000000 max_output_tokens: 128000 + + "gemini-3.5-flash": + context_length: 1048576 + max_output_tokens: 8192 diff --git a/tests/integration/test_stream_invoke_parity.py b/tests/integration/test_stream_invoke_parity.py new file mode 100644 index 00000000..7c01615f --- /dev/null +++ b/tests/integration/test_stream_invoke_parity.py @@ -0,0 +1,104 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +"""Parity test: non-verbose (stream) and verbose (invoke) paths must produce the same CLI-consumed keys. + +The CLI ``scan`` command has two code paths: + - ``--verbose``: uses ``graph.invoke()`` → returns full final state. + - default (non-verbose): uses ``graph.stream()`` and manually accumulates a + subset of keys into a ``result`` dict. + +If the streaming accumulation loop drifts (e.g. a new key is consumed downstream +but never accumulated), the non-verbose path silently produces wrong output. +This test guards against that. +""" + +from __future__ import annotations + +import json +from pathlib import Path + +import pytest + +from skillspector.graph import graph + + +# Keys the CLI reads from the result dict *after* the graph run. +# Derived from cli.py: _write_result, _cleanup_result, exit-code check. +_CLI_CONSUMED_KEYS = frozenset( + { + "report_body", + "sarif_report", + "risk_score", + "temp_dir_for_cleanup", + } +) + + +def _stream_result(state: dict) -> dict: + """Simulate the non-verbose streaming accumulation from cli.py.""" + result: dict = dict(state) + for update in graph.stream(state, stream_mode="updates"): + for _node_name, node_output in update.items(): + if "temp_dir_for_cleanup" in node_output: + result["temp_dir_for_cleanup"] = node_output["temp_dir_for_cleanup"] + if "report_body" in node_output: + result["report_body"] = node_output["report_body"] + if "sarif_report" in node_output: + result["sarif_report"] = node_output["sarif_report"] + if "risk_score" in node_output: + result["risk_score"] = node_output["risk_score"] + return result + + +@pytest.mark.integration +def test_stream_and_invoke_produce_same_cli_keys(tmp_path: Path) -> None: + """Non-verbose (stream) result contains every key that verbose (invoke) produces and the CLI consumes.""" + (tmp_path / "SKILL.md").write_text( + "---\nname: parity-test\n---\n# Safe skill\n", encoding="utf-8" + ) + state: dict = { + "skill_path": str(tmp_path), + "output_format": "json", + "use_llm": False, + } + + invoke_result = graph.invoke(dict(state)) + stream_result = _stream_result(dict(state)) + + # Every key the CLI consumes must be present in *both* results. + for key in _CLI_CONSUMED_KEYS: + assert key in invoke_result, f"invoke result missing CLI key: {key}" + assert key in stream_result, f"stream result missing CLI key: {key}" + + # The actual *values* of the CLI-consumed keys should match (structurally). + # For report_body we compare parsed JSON keys because timestamps differ + # between separate runs. + for key in _CLI_CONSUMED_KEYS: + inv = invoke_result.get(key) + stm = stream_result.get(key) + if key == "report_body": + # Both should parse to JSON with the same top-level keys + inv_parsed = json.loads(inv) + stm_parsed = json.loads(stm) + assert set(inv_parsed.keys()) == set(stm_parsed.keys()), ( + f"report_body top-level keys differ: " + f"invoke={set(inv_parsed.keys())}, stream={set(stm_parsed.keys())}" + ) + else: + assert inv == stm, ( + f"value mismatch for CLI key {key!r}: " + f"invoke={inv!r}, stream={stm!r}" + )