Skip to content

armgit / bla

Public Imported from GitHub
0

πŸ›‘οΈ BLA (Blind LLM Agents) β€” Adversarial defense shield that protects binaries, Electron apps, JS bundles, Python, and WebAssembly against AI reverse-engineering agents like REA. Poison -> Exhaust -> Hallucinate.

2.6 KB
Raw Blame

bla-agents v3.0.0 β€” Agent-Driven Hardening 3.0 (Surgical Weave + Autonomous Remediation)

Date: 2026-11-06

Headline

Major release upgrading BLA into a full Host Agent + MCP Collaborative Hardening system:

  • CrackMe v9 Sovereign (challenge/v9): Polyglot (JS + WASM + HTML + Python) entitlement challenge built through the full 4-Stage BLA 3.0 Playbook (weaveDecoys: true, selfSeal: true, sidecarEntangle: true, weaveScore: 100, integrityCoverageSelfVerified: true, and EXPECTED_SEAL_SHA256 closing the v8 uncovered seal.json sidecar residual).
  • AST-level Surgical Weave Workorders (bla_weave_plan / bla weave-plan): Extracts surgicalWorkorder (entryFunctions, sharedStateSlots, branchCandidates, suggestedIdentifierStems, agentRewriteRecipe), siblingEntanglementCandidates, and the 4-stage agentPlaybook.
  • Structural Weave Verifier (bla_verify_weave / bla verify-weave): Validates Agent-authored source weave edits against all 6 WEAVE_ACCEPTANCE_TESTS (bla-weave-verify/v1), computing weaveScore and returning actionable agentActions for autonomous self-repair.
  • Autonomous Gate & Verification Remediation (bla_lint remediationPlan + bla_verify agentActions): Gives host coding agents exact source-refactoring recipes when packaging gates block or defenseScore soft-caps apply.
  • Skill Guide v3.0.0: Complete 4-Stage Agent-Driven Hardening Playbook (Stage 1 Recon & Gates β†’ Stage 2 Agent Source Surgery β†’ Stage 3 Build & Exhaustion Harden β†’ Stage 4 Equivalence & Weave Verification).
  • Includes post-2.1 hardening: sidecar entry entanglement (sidecar_entangle), cooperative bla_harden progress/cancel, and self-seal missing-sibling hard-fail.

Relative to 2.1.0

Area 2.1.0 3.0.0
bla_weave_plan File-level maturity + static checklist + AST surgicalWorkorder (entryFunctions, sharedStateSlots, branchCandidates, agentRewriteRecipe)
Source weave verification Manual checklist review Automated bla_verify_weave (bla-weave-verify/v1) + weaveScore + agentActions
bla_lint refuse handling Refuse message + allow flag + Structured remediationPlan with sourceRefactorRecipe for Agent source repair
bla_verify feedback Score + gap annotations + Actionable agentActions with recommended MCP tool & args
Skill Guide v2.1.0 one-click guide v3.0.0 4-Stage Agent-Driven Hardening Playbook

Gates

  • npm run build β€” required green
  • npm test β€” required green
  • node dist/cli.js benchmark β€” required green
  • Tags: bla-agents-3.0.0 (and v3.0.0 alias)