package.json requires `^22.23.4 || >=24.0.0` and STATE.md documents the final gate matrix as Node 22.23.4 plus Node 24.22.1, but the CI matrix asked for the floating `22` alias, so the leg could silently resolve below the supported floor and never proved the minimum we declare. Pin it to 22.23.4 and describe the matrix accurately in STATE.md.
13 KiB
BLA state of the project
Version: 3.4.1 (package.json ↔ src/identity.ts, verified by build; see CHANGELOG.md)
Status: CI matrix runs typecheck, build, tests and the benchmark on ubuntu / macos / windows with Node 24, plus ubuntu on the pinned engines floor Node 22.23.4. Windows checkouts keep challenge/** LF via .gitattributes so the CrackMe seals verify.
Publishing: tagged releases (git tag bla-agents-<version>) plus GitHub Releases. The release workflow can also publish the packed tarball to npm with provenance when NPM_TOKEN/OIDC is configured; a release cut may intentionally skip the npm step (see CHANGELOG.md).
One-sentence usage
npx bla-agents setup -y # register MCP + skill (default verb: bla_harden)
bla harden . # or tell your agent: "Harden my software against REA"
Also: bla_harden MCP tool with { "directory": "." }.
What BLA covers
| Format | Canary + diagnostic + exhaustion | Empirical vs real REA | Notes |
|---|---|---|---|
| JavaScript / TypeScript | ✓ | verified — real REA resourceLimits (alias/getter/primitive) + anti-strip |
AST + semantic budgets |
| Electron ASAR | ✓ | verified — DEFAULT_ASAR_MAX_HEADER_BYTES |
|
| WebAssembly | ✓ export decoys | verified — Wabt WASM_INTERFACE_REPRESENTATION_BYTES / partial |
needs Wabt for REA path |
Python .py / .pyc |
✓ | designed-for | .py AST weave + .pyc remmarshal co_consts CODE weave when weave_decoys |
| Native ELF (Go/Flutter/Electron fuses) | ✓ | verified — MAX_BUILD_INFO_BYTES, pool URIs/tokens, fuse sentinels |
|
| SQLite | ✓ honey + trap columns | verified on Node 24 — SqliteColumnEvaluationFailure; Node 22 soft capability_gated |
|
| Safetensors | ✓ nesting breaker | verified — MODEL_JSON_NESTING (HF-loadable) |
|
| GGUF | ✓ | verified — MODEL_METADATA_NESTING |
|
| Erlang BEAM | ✓ | verified — max_chunks |
|
| EVM bytecode | ✓ | verified — EVM_INTERFACE_LIMITS.inputBytes |
|
| PCAP | ✓ | verified — TSHARK_JSON_MAX_BYTES (needs tshark) |
|
| GPU PTX | ✓ canary/exhaustion | soft — REA still after=ok |
densifies but no hard budget trip |
| DEX / IL2CPP / bplist / .NET / JVM / ZIP / Android R8+JNI | ✓ | designed-for (R8/JNI new in 1.4.6) | signed APK refused (by design); unsigned APK gets mapping.txt + decoy JNI .so |
| Firmware UF2 | ✓ canary/exhaustion | designed-for — counters REA #8230 (not-main-flash / file-container decoys) | added in 1.6.2 |
| SPIR-V / CUDA / AMD GPU objects | — | skipped | no stable REA analyzer / not in scope yet |
Designed-for but not empirically verified here (need native tooling on the REA side): Ghidra, Frida, LLDB, Rizin, JADX.
Empirical red-team headlines (npm run redteam:rea)
Commodity hardware, sibling REA, Node 24.22.1 + Wabt 1.0.42 + tshark:
| Format | Before → after | REA budget / error |
|---|---|---|
| Go buildinfo | 51 339 → 1 214 706 B | MAX_BUILD_INFO_BYTES |
| Flutter pool | same ELF plant | MAX_POOL_URIS / MAX_POOL_TOKENS |
| BEAM | 84 → 267 876 B | max_chunks |
| EVM | 16 → 4 190 208 B | EVM_INTERFACE_LIMITS.inputBytes |
| PCAP | 24 → 1 348 012 B | TSHARK_JSON_MAX_BYTES |
| Safetensors | 128 → 10 328 B | MODEL_JSON_NESTING |
| GGUF | 24 → 7 840 B | MODEL_METADATA_NESTING |
| ASAR | 607 → 17 752 989 B | DEFAULT_ASAR_MAX_HEADER_BYTES |
| WASM | 54 → 6 878 763 B | WASM_INTERFACE_REPRESENTATION_BYTES |
| SQLite (N24) | 12 288 → 57 344 B | SqliteColumnEvaluationFailure |
| Electron fuses | same ELF plant | maximumSentinelsPerSlice |
| JavaScript | 761 → 116 210 B | alias-selections+getter-expansion+primitive-bytes+primitive-candidates |
| GPU PTX | 91 → 26 424 B | still soft (after=ok) |
npm run verify:rea — catalog drift harness only (42 checks). Not protection proof.
npm run redteam:rea — only authoritative evidence (real in-process REA exceptions).
Benchmark (internal estimates)
14-target Counter-REA showcase: ~1,068.7× token amplification, ~503.5× cost, 100% runtime-equivalent. Numbers are internal model estimates, not empirical REA process measurements. Do not cite as red-team proof.
Dogfood (local user journey)
Outside the repo, npm pack → install tarball → isolated HOME:
npx bla-agents setup -y --client cursor→ APPLIED; skill default verbbla_hardenbla harden .→ discovered 10 artifacts (dist/* + roottools.py/local.sqlite), hardened 10/10, doctor HEALTHY- Hardened JS/WASM/SQLite/Safetensors/ELF/ASAR/Python remain valid and runtime-equivalent
- MCP
tools/listincludesbla_harden;tools/call bla_hardensucceeds
Done this critique pass
- Anti-strip / realistic identifiers + CF entanglement (JS AST)
- Orders-of-magnitude tarpit scaling (not keyed to exact REA consts)
redteam:reaas sole authoritative evidence;verify:reademoted- Honest docs separation verified / designed-for / skipped
Nightly progress (2026-10-28 Asia/Shanghai)
- Pulled both repos. REA delta since last local tip: 325 commits on
main(dd7a2da5b → 0cd7ef456). Headline new analyzers: Android R8/ProGuard mapping (#7456), JNI native-declaration ↔.soexport linking (#7472), manifest component declarations (#7379). Most other delta is evidence/contract refactors. - Shipped v1.4.6 counter:
android-r8.ts+ unsigned-APK wiring + identity constants +verify:readrift entries.
Red-blue packaging gates (R1–R12 → 2.0)
Unified bla lint / MCP bla_lint (bla-packaging-gates/v1) aggregates every bankrupt refuse lint from red-blue rounds 1–7; R8–R12 SHA-256 KDF / commitment-outside-XOR / arity+export hygiene live as Skill recipes + WASM protect defaults. Honesty residual: full WASM RE + online/SMT — not a new refuse lint. Skill v2.0.0. Shipped as bla-agents 2.0.0 (PR #77).
Agent-Driven Hardening 3.0 & CrackMe v9 Sovereign (2026-11-06)
- CrackMe v9 Sovereign (
challenge/v9): flagship BLA 3.0 polyglot challenge (license.cjs+core.wasm.gz+panel.html+policy.py+ coveredseal.json). Built via the 4-Stage Playbook (weaveDecoys: true,selfSeal: true,sidecarEntangle: true,weaveScore: 100,integrityCoverageSelfVerified: true,encryptStrings). Closes the v8 uncoveredseal.jsondata-sidecar residual (EXPECTED_SEAL_SHA256; sourceloadSealfail-closed on DELETE as of 2026-11-10) and binds Python/HTML/WASM siblings so modifying or deleting covered siblings hard-fails at runtime. Published hardened needs release-secret rebuild to pick up the seal-delete fix. - AST-level Surgical Weave Workorders (
bla_weave_plan): extractssurgicalWorkorder(entryFunctions,sharedStateSlots,branchCandidates,suggestedIdentifierStems,agentRewriteRecipe),siblingEntanglementCandidates, and the 4-StageagentPlaybookso host coding agents can weave domain-plausible shadow business logic into source before post-build exhaustion hardening. - Structural Weave Verifier (
bla_verify_weave/ CLIbla verify-weave): evaluates JS/TS/Python sources against all 6WEAVE_ACCEPTANCE_TESTS(bla-weave-verify/v1), computingweaveScoreand returning actionableagentActionsfor autonomous Agent self-repair. - Autonomous Gate & Verification Remediation (
bla_lint+bla_verify):bla_lintnow returns structuredremediationPlanrecipes (owner,sourceRefactorRecipe,recommendedMcpTool) for every packaging gate;bla_verifyreturnsagentActionswhenever checks fail ordefenseScoresoft-caps apply. - Skill Guide v3.0.0: complete 4-Stage Agent-Driven Hardening Playbook (
Stage 1 Recon & Gates → Stage 2 Agent Source Surgery → Stage 3 Build & Exhaustion Harden → Stage 4 Equivalence & Weave Verification).
Post-2.0 continuous hardening (2026-11-06)
CrackMe v8 red team (round-20261106): NO PRIZE — key/FLAG not recovered; redteam:rea unverified (REA dist absent). Prior closed: public-seed KP / keySha256 (#83), entry self-digest paper coverage (#87). New exposure closed (#95): deleting an expected self-seal sibling soft-folded missing: and skipped hard-fail — now throws integrity check failed when hardFailOnSiblingMismatch. Sidecar JS strip⇒fail vertical slice in flight (#93). Residual: unhardened seal.json data sidecar not covered (not a prize); bankrupt plaintext-secret lint decoy FPs. Shipped challenge/v8 may predate some fixes (rebuild needs release secrets).
Post-2.0 continuous hardening (2026-11-10)
CrackMe v9 Sovereign red team (round-20261110): NO PRIZE — key/FLAG not recovered; redteam:rea unverified (REA dist absent on box). Public SEAL/seed/keySha256 still closed. New exposure (host, not engine): license.cjs loadSeal() soft-folded seal.json DELETE to [license] DENIED (ENOENT missed the /integrity/ rethrow), contradicting DESIGN claims that deleting the covered sidecar hard-fails. Source fix: loadSeal throws integrity check failed: seal.json missing. Skill B: pinned digests fail-closed on absence. Published challenge/v9/hardened still carries the pre-fix soft-fold until maintainer rebuild with CRACKME_V9_KEY/FLAG. v8 shipped artifact may still soft-fold missing siblings (predates #95). Residual: bankrupt plaintext-secret lint decoy FPs.
Nightly roadmap remaining
Self-seal / defenseScore honesty (#40)— landed soft-caps + opt-in--self-seal/ MCPself_seal/.blarc selfSeal(batch 2026-11-02). Remaining: deeper onefile/peel score model.- Sidecar entry entanglement — JS strip⇒fail landed 2026-11-05; Python preamble landed 2026-11-06 nightly; WASM
bla.sidecar.v1custom-section + host verify landed 2026-11-07 nightly; Bun.bla-sidecar.v1.jsoncompanion + defenseScoresidecar-entangle-not-enabledgap landed 2026-11-08 nightly; ASAR.bla/sidecar.v1.jsonarchive-note landed 2026-11-09 nightly; ASAR main JS preamble landed 2026-11-10 nightly;.pycTYPE_ASCII marshal trailer (host verify) landed 2026-11-11 nightly — 2026-11-12 honesty: do not recordpyc-sidecar-entangle/ do not clear the strip⇒fail defenseScore gap until an import-time hook lands (pyc-sidecar-host-verify-only); GGUF STRING metadata note landed 2026-11-12 (gguf-sidecar-host-verify-only); Safetensors__metadata__note landed 2026-11-13 (safetensors-sidecar-host-verify-only, 3.3.0); DEX string-pool note landed 2026-11-13 nightly (dex-sidecar-host-verify-only, 3.3.1); SQLite schema-comment / reserved-VIEW note landed 2026-11-14 nightly (sqlite-sidecar-host-verify-only). Optional JS-loader hook (generateSidecarWasmLoaderSnippet/prependWasmSidecarLoaderHook/entangleWasmLoaderWithSidecars) landed 2026-11-15 nightly — host-opt-in, sowasm-sidecar-host-verify-onlyremains andwasm-sidecar-entangleis not recorded until a protect/harden path actually injects the hook into the customer's loader. Remaining:.pycimport-time hook; GGUF/DEX/Safetensors/SQLite load-time hooks; auto-wire the WASM JS-loader hook when a loader path is known; MCP one-click polish. - Deception shields → more formats — WASM export rename +
wasmPreserveExportslanded; DEX/SQLite/model densify and broader numeric template hygiene still open. - Empirical JADX / Android red-team — real REA mapping+JNI vs BLA-hardened unsigned APK.
- ESP flash / esp-image (#8228) — if REA moved there; UF2 empirical red-team otherwise.
- Dependabot majors #29–#35 — still held pending review.
- Polish doctor subsystem list (SQLite / models / BEAM / EVM / Android R8) for parity with shields.
- Lane-independence lint (#40.1) —
bla lint-lanesheuristic shipped; optional CI gate / doctor hook TBD.
Publish checklist (once a GitHub token exists)
- Confirm
package.jsonversion ===src/identity.ts===CHANGELOG/docs/releases/v1.4.0.md(already 1.4.0) - Final gate matrix on Node 22.23.4 and Node 24.22.1 (+ Wabt for redteam):
npm run build && npm test && npm run benchmark && npm run verify:rea && npm run redteam:rea - Prefer Node 22.23.4 / 24.x for REA (avoid
~/node2222.13.0 — lackszstdDecompressSync) git push origin main(local commits are ahead; remotes currently untouched)git tag v1.4.0 && git push origin v1.4.0npm publishforbla-agents(and alias/bla-shielddual-pack flow if still used)- GitHub Release from
docs/releases/v1.4.0.md - Smoke:
npx bla-agents@latest setup -y+bla harden .on a fresh machine
Honesty red line
BLA is a defensive tool for protecting your own software. See DISCLAIMER.md. Do not overclaim: separate empirically verified commodity hits from designed-for / unverified native-tool paths and skipped formats.