v0.1.4

What's New

Competitive Gap Closure

• Decision memory context injection — last 5 decisions + failure escalation in prompt every turn
• Evidence chain extension — reasoning, source_agent, attempt_number, verification_status on every finding
• 3 webapp tools: sqlmap_parse, xss_detect, jwt_analyze
• 2 pivot/tunnel tools: tunnel_manage, pivot_suggest (BFS path-finding over relationship graph)
• Benchmark infrastructure: bench/verify-claims.ts — 32/32 checks, 5 challenge definitions
• NetExec migration: all crackmapexec references → netexec

Session Critique Fixes (from real Standoff365 testing)

• Mandatory parser workflow — all 9 agent prompts now bind bash commands to parser tools (nmap→nmap_parse, netexec→cme_parse, nuclei→nuclei_parse, etc.). Anti-patterns documented.
• Batch state_update — new batch action accepts {operations: [{action,data},...]} (max 100). Turns 40 sequential LLM steps into 1 for initial engagement setup.
• Context size reduction — toCompactContext() caps: 10 vulns/host, 15 services/host, 30 relationships. OODA fields deduplicated. Conditional injection: full state every 8 turns, summary on others. ~75% token reduction.
• Parallel dispatch strengthening — CORRECT/WRONG examples in orchestrator-mode.txt and pentest.txt

Full Changelog

v0.1.3...v0.1.4