Help us improve
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
By icloudza
Recover cipher algorithms from ARM64 execution traces using the AlgoKiller methodology, with tools for trace analysis, compression, session management, and structured reporting. Includes agents for static analysis offloading and hypothesis auditing.
npx claudepluginhub icloudza/algokiller-plugin --plugin akRun `trace_fold` on the currently-bound trace to produce a block-collapsed derivative trace (typically 99 % compression on hash-loop-heavy traces) and bind to that instead.
Rebind the currently-bound trace into a fresh `<timestamp>/` session directory while keeping the trace path + mode + output_dir unchanged. Useful when starting a new analysis pass without re-typing the trace path.
Show the current algokiller session state — bound trace, mode, output_dir, ledger summary, artifacts written.
Read-only helper that queries Binary Ninja MCP / BinAssistMCP and static-analysis CLIs (radare2, objdump, strings, otool, class-dump) on behalf of the main agent. Spawn this when a static probe will return >1 KB of disassembly, a function decompile, a symbol cross-reference list, or a full S-box / round-constant table — the subagent does the BN/r2 round trips in its own context and hands back named symbols / structural conclusions, not raw listings. The main session never sees 50 KB of disassembly. Use it for tasks like "find xrefs to sm3_compress", "decompile the function at <addr>", "dump the AES S-box from .rodata", "list strings in the .data segment that match `[a-z]+_key_\\d+`".
Independent blue-team reviewer for algokiller Hypothesis Ledger. The main trace-analysis agent MUST spawn this reviewer before any `hypothesis_conclude(final_confidence="high")` call on a load-bearing hypothesis (one that will be cited by the final write_artifact). The reviewer examines the cited hypothesis in isolation — no access to the main agent's reasoning, only the ledger state + raw trace excerpts — and recommends `confirm` / `refute` / `abandon` with reasons. v0.9.1+ — the reviewer records its verdict via `mark_hypothesis_reviewed`, which the server-side conclude(high) gate now requires (FIX
Read-only ledger consistency reviewer. Spawn this near the end of a long ciphertext-recovery / trace-analysis session, BEFORE the main agent calls `write_artifact` on the final `recovered.py` / `report.md`. The curator scans the hypothesis ledger and produces a one-page audit of (a) which concluded hypotheses are cited by name in the planned deliverable, (b) which concluded-but-unreferenced hypotheses should be archived to satisfy the write_artifact bypass-detection gate, (c) any active hypotheses that should be concluded or abandoned before final delivery, (d) any falsification_evidence gaps that would cause `conclude(high)` to fail. This is a sanity pass, not a verdict — the main agent still does the actual conclude/abandon/archive calls.
Read-only helper for the main trace-analysis / ciphertext-recovery agent. Spawn this when a `trace_hexblock` will return >2 KB of bytes_hex (typical for `NSJSONSerialization dataWithJSONObject:` blocks, large memcpy / sprintf outputs, AES key-buffer dumps). The subagent extracts the hexdump in its own context, parses byte ranges into named fields, and returns a compact structured summary — the main session only ever sees the interpretation, not the raw kilobytes of hex. Use it when a `trace_search` hit shows a hexdump block whose length field exceeds 0x800, or when the main session has already burned context on prior hexdumps and another one is queued.
Strong activation entry for AlgoKiller ciphertext-recovery mode. Bind an ARM64 trace, force-load the ciphertext-recovery methodology, and start cipher / algorithm recovery from a target ciphertext.
Strong activation entry for AlgoKiller general trace-analysis mode. Bind an ARM64 trace, force-load the trace-analysis methodology, and answer field semantics / execution flow / detection-point / data-flow questions.
ARM64 trace 通用证据分析方法论。当用户给出一段 ARM64 执行 trace 文件并询问字段含义、执行流、检测点、数据来源、call 边界、buffer 生命周期等开放问题(不属于纯密文还原)时使用。提供搜索键选择、单一目的搜索纪律、call 边界解析、字段语义分层、执行流抽取与检测点分析的完整方法论。激活此 skill 前应已通过 ak MCP 的 bind_trace 工具绑定 trace 文件并选择 mode=general。
Admin access level
Server config contains admin-level keywords
Share bugs, ideas, or general feedback.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
This plugin requires configuration values that are prompted when the plugin is enabled. Sensitive values are stored in your system keychain.
bn_mcp_priorityWhen both Binary Ninja MCPs are online, which to prefer: 'fosdickio' (stdio, GPL-3.0) or 'binassist' (HTTP/SSE, MIT). Leave blank for auto.
${user_config.bn_mcp_priority}default_threadsWorker thread count for trace_constscan / trace_cryptoinstr. Leave blank for auto-detect (host CPU count, capped at 16). Set to 1 for serial scans (debugging) or higher for big multi-GB traces on workstations.
${user_config.default_threads}default_output_dirOptional absolute base directory for analysis reports. Overrides the project-marker walk-up. Equivalent to setting the ALGOKILLER_OUTPUT_DIR env var. Leave blank to use the 5-priority auto-resolution: project-marker walk-up → ~/Documents/AlgoKiller-Reports/.
${user_config.default_output_dir}Claude Code skills for x64dbg debugger automation — state snapshots, memory analysis, and more
Analyze binaries using angr's Python framework. Decompile, recover control flow graphs, find vulnerabilities via symbolic execution, and extract symbols from stripped binaries.
Binary analysis skills for VulHunt MCP - decompilation, dataflow analysis, pattern matching, and more
AI-powered reverse engineering assistant that bridges IDA Pro with language models through MCP.
Binary reverse engineering, malware analysis, firmware security, and software protection research for authorized security research, CTF competitions, and defensive security
Reverse engineer binaries using Ghidra's headless analyzer. Decompile executables, extract functions, strings, symbols, and analyze call graphs without GUI.
语言:中文 | English
面向 ARM64 trace 证据分析与算法/密文还原的 Claude 插件(同时兼容 Claude Code CLI 与 Claude Desktop App)。把 AlgoKiller 方法论打包为 skill,配本地 MCP server 驱动 native ak_search 引擎(14 个 subcommand,专攻 GB 级 trace)。
方法论 + ak_search 引擎原作:AlgoKiller by @lidongyooo 上游贡献
match/context/daemon三个核心子命令(mmap + BMH + 行号索引 + tab 协议 daemon)以及原始方法论 harness。 本仓库在此之上额外扩展了 11 个 native 子命令(regflow/producer/semop/lint/fold/callgraph/modgraph/hexblock/constscan/cryptoinstr/bytes,详见 tools/search/README.md)并把整套打包为 Claude 插件(Claude Code / Desktop 都能加载)。原始代码版权归上游作者;plugin 自身的扩展代码 MIT。
claude plugin marketplace add icloudza/algokiller-plugin
claude plugin install ak@ak-suite
更新:
claude plugin marketplace update
claude plugin update ak@ak-suite
Claude Code REPL 里也可以用
/plugin marketplace add ...//plugin install ...slash 等价命令。手动安装方式见下方 完整安装说明。
Cursor / Codex:本仓库也提供标准 stdio MCP server 配置示例。Cursor 可直接使用 .cursor/mcp.json;Codex 可复制 examples/mcp/codex.config.toml 到 ~/.codex/config.toml。详见 Cursor and Codex MCP setup。注意:非 Claude 客户端只能直接使用 MCP 工具,不会自动获得 Claude slash commands / skill 自动加载。
ak:ciphertext-recovery —— 密文 / header / token 反向还原加密、签名、编码算法ak:trace-analysis —— 字段语义 / 执行流 / 检测点 / buffer 生命周期等开放问题/ak: 命名空间)
/ak:ciphertext <trace> <task> —— 强激活:密文 / 算法还原模式/ak:general <trace> <task> —— 强激活:通用 trace 分析模式/ak:status —— 当前 session digest(绑定 / output_dir / ledger 摘要 / artifacts)/ak:rebind —— 同 trace 新 <timestamp>/ 重绑(多次分析对照不冲突)/ak:fold —— 折叠当前 trace(hash loop 类 99% 压缩,然后可选 rebind 到折叠版)bind_trace / pick_output_dir(弹原生目录选择器)/ write_artifact / list_artifacts / read_artifacttrace_search / trace_contexttrace_regflow(寄存器演化)/ trace_producer(找值的最近写入者)/ trace_semop(指令语义分类,11 类)trace_lint(一遍 JSON 体检)/ trace_fold(block-aware 折叠,115 MB → 1.1 MB)trace_callgraph(Top-K / xref)/ trace_modgraph(跨模块矩阵)/ trace_hexblock(call+args+hexdump+ret 结构化)trace_constscan(97 个 hash/cipher/ecc/crc/mac 常数指纹 — 95 个 scalar literal + 2 个 NEON SIMD 广播;含 MD5 init+T 表 / SHA-256 init+K 表 / SM3 init+T_j / SHA-3 / CRC32 / AES sbox+Te0 / SM4 / ChaCha20 / Poly1305 / SipHash / HMAC ipad-opad (scalar + SIMD) / P-256 / secp256k1 / Ed25519 / Curve25519;带 verdict 分级 real / real_simd / weak / alu_only;MD5.T[i] 等主循环常数附 block_count_estimate)/ trace_cryptoinstr(ARM Crypto Extensions 硬件指令:AES/SHA-1/SHA-256/SHA-512/SHA-3/SM3/SM4/GHASH)/ trace_bytes(hex 字面量含自动反序变体)run_static_tool —— 白名单调用系统 CLI(radare2 / binutils / LLVM / jtool2 / class-dump / ripgrep / jq)discipline_reminder;每 20 次附 discipline_full_reinjection 完整规则段hypothesis-reviewer —— 独立 context 蓝军,hypothesis_conclude(high) 之前主 agent spawn 它做独立证据审查。详见 docs/agents.md。trace-hexdump-extractor —— 大 hexdump 在子上下文里解析成结构化字段返回,主会话不被 raw bytes 淹没binary-static-inspector —— 包 Binary Ninja MCP / BinAssistMCP / run_static_tool,主会话只接收符号 / decompile 结论而非 50 KB disassemblyledger-curator —— 写交付物前的 ledger 一致性 audit,识别未引用的 concluded 假设PreCompact(auto) —— 长 scan 持锁中阻断 auto-compact + dump session 摘要到磁盘PreCompact(manual) —— 用户主动 /compact 时不阻断,仍 dump 摘要SessionStart(compact) —— compact 后注入 ledger 摘要 + [H<n>] 引用规则SessionStart(startup|resume) —— 自动装 pyright(驱动 .lsp.json)+ 环境诊断Stop —— 写 session-summary.mdSubagentStop —— 校验 hypothesis-reviewer 真的落了 mark_hypothesis_reviewedPreToolUse(write_artifact) —— 客户端预警:draft 里 [H<n>] 引用数 < ledger concluded 数时 emit warningtrace_constscan / trace_cryptoinstr 自动按 CPU 数据并行(默认 = 主机核数,封顶 16,可经 threads 参数覆盖)。4.5 GB / 48 M 行 trace 上 constscan 8 线程 ≈ 19 s(单线程 121 s),输出对所有线程数 byte-identical。~/.algokiller/active-scans.lock 持 kernel-flock;PreCompact 钩子据此阻断 auto-compact,进程崩溃时内核自动释放锁,无 stale-lock 风险。bin/ak_search 按本机架构编译)二进制架构不匹配时重编:
cd tools/search && make
cp ak_search ../../server/bin/ak_search
一行装机命令见顶部 🚀 快速安装。
手动安装(备选):克隆仓库后
+ → Plugins → Add plugin → 选本地目录claude plugin install .(或把仓库路径加进 claude plugin marketplace add <local>)git clone https://github.com/icloudza/algokiller-plugin