Skip to content

feat(security): add Security Guard module for attack surface defense#111

Open
MicroMilo wants to merge 2 commits into
OpenBMB:mainfrom
MicroMilo:main
Open

feat(security): add Security Guard module for attack surface defense#111
MicroMilo wants to merge 2 commits into
OpenBMB:mainfrom
MicroMilo:main

Conversation

@MicroMilo
Copy link
Copy Markdown

@MicroMilo MicroMilo commented Jun 2, 2026

添加安全防护,围绕 posthook 展开,详见 commit message

MicroMilo added 2 commits June 2, 2026 16:58
Add a non-blocking security defense layer protecting 4 attack surfaces:

**MCP instruction sanitization** — 4-level pipeline (XML escape → length
truncation → pattern detection → warning injection) applied to MCP server
instructions injected into the system prompt.

**MCP tool output guard** — PostToolUse hook scanning mcp__* tool outputs
against configurable suspicious patterns (curl, eval, bash -c, etc.).

**Web fetch guard** — PreToolUse guard wrapping web_fetch content with
boundary markers to prevent prompt injection from external pages.

**Annotation forgery guard** — PreToolUse guard detecting MCP tools that
declare read-only annotations but have names/params suggesting mutating
operations.

Also adds:
- User-configurable SecurityPolicy with deep merge (objects recurse,
  arrays concat) in ~/.pilotdeck/security-policy.json
- Auto-generates commented policy template on first access
- Credential detection in bash output via hook guard
- Dynamic MCP instruction merging in PluginRuntimeExtensionResolver
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant