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.

3.2 KB
Raw Blame

CrackMe v8 β€” Aegis Design

Status: successor to CrackMe v7 Meridian. Built with bla-agents 2.0 after red/blue rounds 1–12; the published tree is hardenProject() output.

Product framing

Aegis is a small entitlement unlock:

Command Behavior
node license.cjs <16-char-key> Unlock certificate / FLAG on success; uniform DENIED otherwise

There is no in-browser panel derive. panel.html is operator UI only.

Why not re-harden Meridian v7

Meridian cleartext + gen.mjs live in a maintainer-private store (only hardened + SEAL are in git). Round-1 already recorded that an honest 2.0 re-harden of v7 is impossible without that store. Aegis is the public graduation of the standing red/blue entitlement design instead.

Recipes absorbed from R1–R12

Round Lesson Aegis response
R1 Stable WASM ABI + independent 1-byte lanes No wasmPreserveExports / no lane tables; bankrupt rename default
R2–R4 Reversible seals, JS AES unwrap, panel oracles, soft self-seal One-way SHA-256 seal; no AES unwrap; no panel derive; --self-seal
R5–R6 Hydrate dump / XOR-mask sidecar Unlock stays in WASM; no mask sidecar
R7–R9 Seal.revision XOR-KP + weak mask KDF + structure KP SHA-256(key) mask KDF; commitment outside XOR; seed-derived rounds+pad3
v8 red Public SEAL seed β†’ schedule KP; public keySha256 β†’ offline key oracle Harden/schedule seed private (derived from release secrets), SEAL v2 publishes only label + flagSha256; bla_lint warns public-materialization-seed / public-key-digest-oracle
R10–R12 Arity-2 alias stack / export-order / body idx 0 Prefer arity-2 aliases; export-order shuffle; unlock body-index scatter
P2 Full WASM RE + SMT Honest residual β€” not refused as a packaging lint

Architecture

challenge/v8/hardened/
  license.cjs      ← hardened host gate (arity + FLAG + certificateSha256)
  panel.html       ← hardened operator UI (no token derive)
  core.wasm.gz     ← unlock-in-WASM (gzipped; ~7 MB raw)
  seal.json        ← runtime one-way seal (revision + certificateSha256)
challenge/v8/SEAL.json  ← public digests / knobs only

Unlock (happy path)

  1. Host loads hardened core.wasm, finds arity-2 exports (order/name-independent).
  2. Invokes candidates; accepts UTF-8 matching FLAG{…}.
  3. Requires SHA-256(result) === seal.certificateSha256.
  4. Inside WASM: mask = SHA-256(key[16]), unmask pack salt||rounds||pad3||entropyPad[32]||clen||cipher, compare clear commitment segment to SHA-256(ARX_state), decrypt certificate.

Wrong key / honeypot canary / missing sibling β†’ uniform deny (honeypots never call WASM unlock).

Honest scope

Solvable by a determined reverse engineer who recovers the in-WASM unlock (SHA-256 KDF + ARX + commitment layout) and searches the key space β€” not by packaging footguns that R1–R12 closed. Gates do not claim white-box secrecy against full WASM RE / online SMT.

Maintainer recipe

npm run build
CRACKME_V8_KEY='…' CRACKME_V8_FLAG='FLAG{…}' npm run crackme:v8:build
npm run crackme:v8:seal
# fixture-only CI path (no release secrets):
npm run crackme:v8:build -- --fixture