PenMaster v0.3.0 - 21 Tool Autonomous Pentest Agent

What's Changed

• Feature/logging replay system by @XenoCoreGiger31 in #24
• feat: add memory system with short-term buffer, JSONL logs, and episo… by @XenoCoreGiger31 in #25

New Contributors

• @XenoCoreGiger31 made their first contribution in #24

Full Changelog: SOCIAL_ENGINEERING_TOOLKIT...v0.3.0

ADDED_SOCIAL-ENGINEERING-TOOLKIT

Full Changelog: https://github.com/XenoCoreGiger31/Local-Model/commits/v1.2

Added Social Engineering Toolkit (SET) integration to the PenMaster Security agent suite. The agent can now autonomously launch SET-based attacks including credential harvesting and site cloning directly through the MCP tool interface. This brings PenMaster's total autonomous toolset to 18 tools covering reconnaissance, exploitation, brute force, and now social engineering attack vectors.

"Autonomous Tool Reasoning: MCP Agent - Loop Security - Kali Linux VM Release"

Release Notes — Sovereign Agent v1 (Negative Experience Cache Update)

• 🧠 Introduced Persistent Negative Experience Cache (Sovereign Agent Layer v1)
  Cross-session memory system that tracks failed tool executions and prevents repeating ineffective actions.
• 🚫 Added failure fingerprinting system
  Identical tool attempts are hashed and recognized across sessions for consistent behavior control.
• ☠️ Implemented 2-strike permanent block logic
  Failed tool patterns are retried once, then automatically blacklisted to reduce wasted execution cycles.
• 🔁 Integrated cache-aware execution gate in agent_loop
  All tool steps now pass through should_attempt() before execution.
• 📉 Added failure recording + recovery handling
  Failed attempts are logged with reason tracking; successful retries clear outdated negative memory.
• 🧾 Enhanced observability
  Cache stats, dumps, and logging improvements for debugging agent decision history.
• ⚙️ Improved attack loop efficiency
  Reduced redundant tool calls and improved decision routing in iterative engagement cycles.

Persistent_Neg_Experience_Cache_Sovereign_Agent_v1

Full Changelog: v1.1-animated-banner...Persistent_Negative_Experience_Cache_v1

🧠 Persistent Negative Experience Cache (Sovereign Agent Layer v1)

The Persistent Negative Experience Cache gives the agent long-term memory of failed actions across sessions. Each tool execution is fingerprinted and tracked, allowing the agent to recognize previously unsuccessful approaches and avoid repeating them indefinitely.

Failed actions receive one retry opportunity. After two confirmed failures, the action is permanently blocked and automatically skipped in future engagements. If a previously blocked action later succeeds, its failure record is cleared and normal operation resumes.

This update improves efficiency, reduces redundant tool execution, and enables more adaptive autonomous decision-making through persistent experience-based learning.

PenMaster Security — Animated Brand Banner

PenMaster Security Animated Brand Banner

Animated version of the PenMaster Security brand banner.
Replaces the static SVG previously used in the README.

"Autonomous Tool Reasoning: MCP Agent - Loop Security - Kali Linux VM Release"

Full Changelog: https://github.com/XenoCoreGiger31/Local-Model/commits/AtonomousSecurityagentWhat’s Changed
What’s Changed

•	agent_loop.py — Fixed false positive exploit detection logic
•	mcp_server.py — Fixed false positive exploit detections
•	.gitignore — Updated to reflect false positive exploit fix patches

Bug Fixes

•	Resolved ncrack users.txt bug
•	Fixed Hydra flag handling issue

Notes

Core stability patch for the autonomous pentesting agent loop.
Improves accuracy of exploit identification against live targets.