feat: enhance red team automation with exploitation toolsets and forced exploitation logic

[l50]

Key Changes:

• Added new toolsets for coercion, MSSQL, ACL abuse, CVE exploits, trust attacks, and lateral movement
• Introduced a vulnerability discovery hook that forces exploitation when vulnerabilities are found
• Overhauled BloodHound and Certipy tools to provide structured, actionable outputs
• Updated agent instructions and system prompt to require exploitation of all discovered paths before summarizing

Added:

• New toolsets for exploitation, including:
  • CoercionTools (PetitPotam, Coercer) for authentication coercion attacks
  • MSSQLTools for SQL Server exploitation and impersonation chains
  • ACLExploitTools for exploiting Active Directory ACL misconfigurations (pywhisker, bloodyAD)
  • CVEExploitTools for known vulnerabilities (noPac, PrintNightmare)
  • TrustAttackTools for cross-domain/forest escalation (raiseChild)
  • LateralMovementTools for remote access (evil-winrm, psexec)
• Vulnerability discovery hook that monitors tool output and injects guidance to immediately exploit discovered vulnerabilities (e.g., ESC1, ACL abuse, delegation, MSSQL impersonation)
• Structured output parsing for BloodHound and Certipy tools, including JSON summaries of attack paths, vulnerable templates, and prioritized recommended actions
• Expanded import and export logic in ares.tools.red.__init__.py to register all new toolsets

Changed:

• BloodHound toolset now parses output for actionable attack paths, ACL abuse, delegation, and high-value targets, returning structured data and step-by-step recommendations
• Certipy toolset parses ADCS enumeration output for ESC vulnerabilities and provides structured exploitation steps and parameters
• create_redteam_agent now includes all new exploitation toolsets and registers the vulnerability discovery hook to enforce exploitation-first workflows
• System instructions for red team agents now explicitly forbid summarizing operations until all discovered vulnerabilities/paths have been exploited, with detailed anti-summarization and fallback chain rules
• Example commands and error recovery logic included in system prompt to guide agent behavior

Removed:

• Redundant static tool references and outdated instructions, consolidating all tool import/export logic to support new exploitation automation
• Legacy examples in docstrings that did not reflect new exploitation toolsets or workflows

[linear]

CAP-838 Fix Red Agent: Behavioral Gaps & Missing Tool Wrappers

Description:
The Red Agent currently fails to exploit most discovered vulnerabilities due to behavioral shortcomings and lack of tool wrappers for key attack techniques. This task aims to address these gaps by improving the agent’s exploitation behavior and exposing installed tools through new wrappers, thereby enabling comprehensive automated exploitation.

---

Objective:

Ensure the Red Agent reliably exploits all discovered attack paths by introducing structured output parsing, proactive exploitation triggers, anti-summarization directives, and wrappers for critical exploitation tools.

---

Scope of Work:

• Refactor discovery tools (e.g., certipy_find, BloodHound) to parse output and return structured, actionable data
• Implement a vulnerability discovery hook to trigger immediate exploitation actions when vulnerabilities are detected
• Update system instructions to include an anti-summarization directive and require exploitation before summary or completion
• Add tool wrappers for installed binaries: petitpotam, coercer, mssqlclient.py, pywhisker, bloodyAD, noPac, PrintNightmare, raiseChild.py, evil-winrm
• Test new wrappers to ensure agent can invoke and process output from each tool
• (Optional, P1) Implement a periodic priority check hook to remind the agent of unexploited discoveries

---

Dependencies:

• Access to source code for src/ares/tools/red/network.py and related agent logic
• Existing installations of exploitation tools (as listed)
• Ability to update agent system instructions/templates
• Test environment with representative AD infrastructure

---

Acceptance Criteria:

1. Discovery tools (certipy_find, BloodHound) return structured output with explicit exploit parameters (e.g., CA/template names, attack path details)
2. When a vulnerability is found, the agent immediately attempts exploitation without manual intervention
3. Agent does not prematurely summarize or complete operations until all exploitable paths have been attempted
4. Wrappers for all listed tools are implemented and accessible to the agent via @dn.tool_method
5. Agent successfully invokes at least one exploitation command for each new wrapper in end-to-end tests
6. (If implemented) The agent receives periodic reminders to exploit all discovered vulnerabilities

---

Additional Notes:

• Prioritize behavioral fixes (structured output, exploitation triggers, anti-summarization) before adding tool wrappers for maximum impact
• Reference implementation priority and quick wins section for incremental delivery
• Test with varied AD environments to ensure robust parsing and exploitation logic
• For structured output, consider JSON or dicts to encapsulate actionable findings
• See logs and gap analysis in the original plan for specific examples of behavioral failure

---