mirror of
https://github.com/agent0ai/agent-zero.git
synced 2026-05-24 22:04:03 +00:00
Calibrate scanner prompts around demonstrated risk instead of the mere presence of common plugin capabilities. Treat scoped credentials, network calls, filesystem access, subprocesses, prompts, and generated assets as expected behavior when they match the declared plugin purpose, while keeping warnings and failures for ambiguity, unsafe handling, concealment, exploitability, or purpose mismatch. Add regression coverage for the rendered scanner prompt so this calibration is preserved. |
||
|---|---|---|
| .. | ||
| api | ||
| extensions/webui/confirm_dialog_after_render | ||
| helpers | ||
| webui | ||
| plugin.yaml | ||
| README.md | ||
Plugin Scanner
Run an LLM-guided security review of third-party Agent Zero plugins from a Git repository.
What It Does
This plugin builds a structured scanning prompt from a selectable checklist, runs that prompt in a temporary agent context, and returns a markdown report describing the plugin's security posture.
Main Behavior
- Prompt-driven scan
- Loads scan checks and a markdown prompt template from the plugin's
webui/assets.
- Loads scan checks and a markdown prompt template from the plugin's
- Temporary scan context
- Creates a temporary chat context, logs the generated prompt into it, starts the agent immediately, and waits for the model result.
- Parallel-friendly execution
- Each scan runs in its own chat context; the plugin does not serialize scans behind a "wait for another scan" queue.
- Selectable checks
- Supports scanning all checks by default or only the subset selected by the caller.
- UI integration
- Includes API endpoints and web UI files for logging the prompt, starting the scan, and running scans synchronously.
Key Files
- Scan runner
api/plugin_scan_run.pyperforms a synchronous end-to-end scan and returns the report.
- Prompt builder
helpers/prompt.pyloads check definitions and renders the final scan prompt.
- Additional APIs
api/plugin_scan_queue.pylogs the prompt into the temporary chat.api/plugin_scan_start.pystarts the agent in that chat.
Configuration Scope
- Settings sections: none
- Per-project config:
false - Per-agent config:
false
Plugin Metadata
- Name:
_plugin_scan - Title:
Plugin Scanner - Description: Security scanner for third-party A0 plugins.