securevibes-scanner

在代码库上运行人工智能驱动的应用程序安全扫描。当要求扫描代码中的安全漏洞、生成威胁模型、检查代码中的安全问题、运行增量安全扫描或通过 cron 设置持续安全监控时使用。支持完整扫描（一次性）和增量扫描（cron 驱动，仅新提交）。

## 概述（中文）

在代码库上运行人工智能驱动的应用程序安全扫描。当要求扫描代码中的安全漏洞、生成威胁模型、检查代码中的安全问题、运行增量安全扫描或通过 cron 设置持续安全监控时使用。支持完整扫描（一次性）和增量扫描（cron 驱动，仅新提交）。

## 原文

# SecureVibes Scanner

AI-native security platform that detects vulnerabilities using Claude AI. Multi-subagent pipeline: assessment → threat modeling → code review → report generation → optional DAST. Supports incremental scanning for continuous monitoring.

## Prerequisites

1. Install the CLI: `pipx install securevibes` (preferred) or `uv tool install securevibes`. Avoid `pip install` — it can create stale shims if you have multiple Python environments.
2. Authenticate with Anthropic (one of):
   - **Max/Pro subscription (recommended):** If you're authenticated via Claude Code or Claude CLI OAuth, no API key is needed. The Claude Agent SDK picks up your OAuth session automatically. When running inside OpenClaw, leave `ANTHROPIC_API_KEY` unset or blank — the SDK handles auth.
   - **API key:** `export ANTHROPIC_API_KEY=your-key-here` (from console.anthropic.com)

## Security Notes

- Always use the `scripts/scan.sh` wrapper for full scans — it validates paths and rejects shell metacharacters before invoking `securevibes`.
- **Never interpolate unsanitized user input into shell commands.**
- The wrapper uses `realpath` to resolve paths safely and rejects any path containing `;`, `|`, `&`, `$`, backticks, or other metacharacters.
- **Scan targets must be local directories.** Clone remote repos to a known safe location first, then pass the resolved path to the wrapper.
- **DAST scans make network requests** to the `--target-url` you provide. Only use against apps you own or have permission to test.

## Execution Model

**Full scans take 10-30 minutes across 4 phases.** Run them as background jobs (cron or subagent), not inline.

**Incremental scans take 2-10 minutes** — they only scan commits since the last run.

## Full Scan (One-Shot)

### Running a Scan

1. Clone the target repo to a local directory
2. Run the wrapper script: `bash scripts/scan.sh /path/to/repo --force --debug`
3. Results appear in `/path/to/repo/.securevibes/`

### Background Execution (Recommended)

For OpenClaw users, schedule scans as cron jobs:

- Use `sessionTarget: "isolated"` with `payload.kind: "agentTurn"`
- Set `payload.timeoutSeconds: 2700` (45 minutes) to allow all phases to complete
- Use `delivery.mode: "announce"` to get notified when done

The agentTurn message should instruct the subagent to:
1. `cd` into the repo and `git pull` for latest code
2. Clean previous `.securevibes/` artifacts
3. Run `securevibes scan . --force` via the wrapper script
4. Read and summarize the results from `.securevibes/scan_report.md`

## Incremental Scan (Continuous Monitoring)

The incremental scanner (`ops/incremental_scan.py`) tracks the last-scanned commit and only scans new commits. Designed for cron-driven continuous security monitoring.

### How It Works

1. Tracks an anchor commit in `.securevibes/incremental_state.json`
2. On each run: fetches remote, compares HEAD to anchor
3. If new commits exist: runs `securevibes pr-review` on the diff
4. Updates anchor to new HEAD after successful scan
5. If no new commits: exits cleanly (no scan, no cost)

### Setup

#### Step 1: Run an initial full scan (if not already done)

The incremental scanner requires `.securevibes/SECURITY.md` and `.securevibes/THREAT_MODEL.json` to exist. These come from an initial full scan:

```bash
securevibes scan <repo-path> --model sonnet
```

Skip this step if the repo already has a `.securevibes/` directory with these files.

#### Step 2: Bootstrap incremental state

Run the wrapper once to seed the anchor commit (no scan runs, just records current HEAD):

```bash
python3 ops/incremental_scan.py --repo <repo-path> --remote origin --branch main
```

This creates `.securevibes/incremental_state.json` with `status: "bootstrap"`.

#### Step 3: Configure the cron

For OpenClaw users, create a cron job:

```bash
openclaw cron create \
  --name "securevibes-incremental" \
  --cron "*/30 * * * *" \
  --tz "America/Los_Angeles" \
  --agent main \
  --session isolated \
  --timeout-seconds 900 \
  --announce \
  --message "Run incremental security scan: python3 <skill-path>/ops/incremental_scan.py --repo <repo-path> --remote origin --branch main --model sonnet --severity medium --scan-timeout-seconds 600. Read .securevibes/incremental_scan.log for results. If new findings, summarize them."
```

Replace `<skill-path>` with the installed skill path and `<repo-path>` with the target repo.

#### Step 4: Verify

```bash
# Check state
cat <repo-path>/.securevibes/incremental_state.json

# After first scheduled run, check logs
tail -10 <repo-path>/.securevibes/incremental_scan.log

# Check findings
cat <repo-path>/.securevibes/PR_VULNERABILITIES.json
```

### Incremental Scanner Options

```
python3 ops/incremental_scan.py [options]
```

| Option | Description |
|--------|-------------|
| `--repo` | Repository path (default: `.`) |
| `--branch` | Branch to track (default: `main`) |
| `--remote` | Git remote (default: `origin`) |
| `--model` | Claude model: `sonnet`, `haiku` (default: `sonnet`) |
| `--severity` | Minimum severity: `critical`, `high`, `medium`, `low` |
| `--scan-timeout-seconds` | Timeout per scan command (default: `900`) |
| `--git-timeout-seconds` | Timeout for git operations (default: `60`) |
| `--rewrite-policy` | History rewrite handling: `reset_warn`, `strict_fail`, `since_date` |
| `--since` | Override: scan commits since this date (ISO or YYYY-MM-DD) |

### Operational Guarantees

- **File lock** at `.securevibes/.incremental_scan.lock` prevents overlapping runs
- **Atomic state writes** (`fsync` + `os.replace`) prevent corruption
- **Structured logging** at `.securevibes/incremental_scan.log`
- **Run records** saved to `.securevibes/incremental_runs/` (one JSON per run)

### Rewrite Policy

When `last_seen_sha` is not an ancestor of the new remote HEAD (e.g., force push):

| Policy | Behavior |
|--------|----------|
| `reset_warn` | Reset anchor to new HEAD, continue |
| `strict_fail` | Fail and keep current anchor |
| `since_date` | Run a `--since <today>` scan for visibility, keep previous anchor |

## Full Scan Commands Reference

### Scan

`securevibes scan <path> [options]`

| Option | Description |
|--------|-------------|
| `-f, --format` | `markdown` (default), `json`, `text`, `table` |
| `-o, --output` | Custom output path |
| `-s, --severity` | Filter: `critical`, `high`, `medium`, `low` |
| `-m, --model` | Claude model (e.g., `sonnet`, `haiku`) |
| `--subagent` | Run one phase: `assessment`, `threat-modeling`, `code-review`, `report-generator`, `dast` |
| `--resume-from` | Resume from a specific phase onwards |
| `--dast` | Enable dynamic testing (requires `--target-url`) |
| `--target-url` | URL for DAST (e.g., `http://localhost:3000`) |
| `--force` | Skip prompts, overwrite existing artifacts |
| `--quiet` | Minimal output |
| `--debug` | Verbose diagnostics |

### Report

`securevibes report <path>` — Display a previously saved scan report.

## Mapping Requests to Actions

| User Says | Action |
|-----------|--------|
| "Scan this for security issues" | Full scan: `bash scripts/scan.sh <path> --force` |
| "Quick security check" | Full scan: `bash scripts/scan.sh <path> -m haiku --force` |
| "Threat model this project" | `bash scripts/scan.sh <path> --subagent threat-modeling --force` |
| "Just review the code" | `bash scripts/scan.sh <path> --subagent code-review --force` |
| "Show only critical/high findings" | `bash scripts/scan.sh <path> -s high --force` |
| "Full audit with DAST" | `bash scripts/scan.sh <path> --dast --target-url <url> --force` |
| "Set up continuous scanning" | Incremental setup: Steps 1-4 above |
| "Monitor this repo for security issues" | Incremental setup: Steps 1-4 above |
| "Show last scan results" | `securevibes report <path>` |

## Subagent Pipeline

Runs sequentially. Each phase builds on the previous:

1. **assessmen