[Rules] Add TG-AGENT-* rule family for Agentic AI prompt injection and MCP tool scoping

[githubmofo]

Background

Agentic AI repositories (such as Hermes Agent, AutoPentest-AI, or MCP servers) introduce attack surfaces not covered by traditional web application rule families.

Proposed Rules (Tracked for v0.7.1)

• TG-AGENT-001: Direct/Indirect Prompt Injection in System Context Files.
• TG-AGENT-002: Unsafe Tool Dispatch & Shell Execution without Sandboxing.
• TG-AGENT-003: Overly Broad MCP Tool Scoping & Credential Access.
• TG-AGENT-004: Persistent Memory & Cross-Session Information Leakage.

Acceptance Criteria

• Add rule documentation files under rules/agent/.
• Add paired vulnerable and hardened agentic fixtures under tests/fixtures/agent/.
• Incorporate rules into runner.py catalog verification.

[githubmofo]

Resolved in v0.9.2: Added the complete TG-AGENT-* rule family under rules/agent/:

• TG-AGENT-001: Prompt Injection in System Context Files
• TG-AGENT-002: Unsafe Tool Dispatch & Shell Execution without Sandboxing
• TG-AGENT-003: Overly Broad MCP Tool Scoping & Credential Access
• TG-AGENT-004: Persistent Memory & Cross-Session Information Leakage
  All rules verified through runner.py catalog integrity.