CipherSwarm Phase 5: Master Summary#

Table of Contents#

• CipherSwarm Phase 5: Master Summary
  • Table of Contents
  • Mission
  • DAG-based Cracking Campaigns
    • Why a DAG?
  • CORE PILLARS
    • Advanced Task Scheduling Advanced Task Scheduler
    • Agent Sync + Health Framework Agent Sync Extensions
    • Agent Collaboration Model Agent-Server Collaboration Vision
    • Hard Password Attack Intelligence Hard Password Attack Strategies
  • INTEGRATION THEMES
    • Intelligence-Driven Strategy
    • Agent Self-Governance
    • Fine-Grained Orchestration
    • Observability & Learning
  • Immediate Implementation Tracks
  • Suggested Skirmish Sequence

Mission#

Phase 5 transforms CipherSwarm from a high-performance orchestrator into an adaptive, feedback-driven cracking intelligence system, with smarter agents, real-time scheduling, dynamic planning, and strategic attack optimization.

DAG-based Cracking Campaigns#

This is the CipherSwarm's effort to implement Directed Acyclic Graph (DAG) based cracking campaigns. In CipherSwarm’s context, a DAG is used to model the flow of attack phases — where each node represents a cracking strategy (e.g., dictionary+rule, mask, brute-force), and arrows show which phases logically follow others.

Why a DAG?#

• Directed: Attack steps move forward — you don’t re-run prior phases unless explicitly configured.
• Acyclic: No loops — each path flows from start to completion without circling back.
• Graph: Multiple branches can exist in parallel, enabling exploratory or fallback strategies.

CORE PILLARS#

Advanced Task Scheduling Advanced Task Scheduler#

• WorkSlice + TaskPlan system based on precomputed keyspace divisions
• Fully supports hybrid, mask, brute-force, and incremental attack types
• Real-time slice leasing with reclaim logic and keyspace coverage tracking
• Agent scoring considers hashrate benchmarks, throttling, and uptime
• Supports crackless watchdogs, thermal-aware scoring, and background task prioritization

Agent Sync + Health Framework Agent Sync Extensions#

• Backoff Signals: Agents are explicitly told to pause based on system or agent health
• Load Smoothing: Randomized heartbeats and sync intervals to prevent traffic spikes
• Failure Pattern Tracking: Rolling agent reliability scores impact task assignment
• Lease Expiry & Reclaim: TTL-based Redis tracking for automatic task reclamation
• Agent Local Heuristics: Agents throttle themselves based on temp, load, or guessrate

Agent Collaboration Model Agent-Server Collaboration Vision#

• Structured Status Streaming via /status with --status-json parsing
• Self-Tuning Agents adjust workload profile (-w) and runtime params dynamically
• Offline Recovery: Agents checkpoint slice metadata mid-task; server accepts partial completion
• Live Plan Adjustment: Server adapts slice sizing and prioritization in real time
• Crack Feedback Loop: Successful cracks influence dictionary/rule/mask strategy
• Performance History: Server models agent capabilities per hash type
• Capability Signaling: Agents report hash type support, memory, load
• Optional: Agent Karma, Slice Replay, DAG Auto-Growth

Hard Password Attack Intelligence Hard Password Attack Strategies#

• Dynamic Wordlists: Meta-wordlists, frequency sorting, crack-informed candidates

• Rule Learning & Debug Parsing: Derive rules from cracked pairs and --debug-mode=3 - See Learned Rules Parser Plan

• Markov Modeling: Automatic hcstat2 generation per project; opt-in UI toggle - See Markov Auto-Generation Plan

• PACK-Inspired Intelligence:

  • Internal maskgen, rulegen, statsgen, policygen clones

• Graph-Driven Campaigns: DAG-style phased attack planning

• LLM/Trigram Expansion: AI-inspired password candidate generation

• Advanced DAG Logic:

  • Crack origin attribution
  • Entropy bucketing
  • DAG trimming or extension
  • Agent-affinity weighting
  • Hot slice promotion

INTEGRATION THEMES#

Intelligence-Driven Strategy#

• Project-aware wordlists, rules, and mask evolution
• Feedback from cracks, rejects, and agent behavior
• Environment-specific password morphology modeling

Agent Self-Governance#

• Dynamic resource tuning
• Autonomy in edge cases (overheating, reboots)
• Participation in planning via capability reporting and crack insight

Fine-Grained Orchestration#

• Per-slice telemetry, live reassignment, and execution tracing
• Fault-tolerant leasing
• DAG-based campaign modeling
• Background vs. primary task scheduling

Observability & Learning#

• Per-campaign attribution
• Rule effectiveness graphs
• Cracked-password pattern clustering
• Slice replay for debug/forensics

Immediate Implementation Tracks#

Suggested Skirmish Sequence#

1. ✅ Extend TaskPlan + WorkSlice model
2. ✅ Implement lease tracking + reclaim worker
3. ✅ Implement Markov stats + hcstat2 autogen
4. ✅ Add debug rule parser → learned.rules
5. ✅ Build /status endpoint + log pipeline
6. ✅ Add PACK-core modules: mask, rule, stats
7. ✅ Add DAG node scoring + trigger logic