find-crypto-entry

스타9

포크1

업데이트2026년 7월 12일 05:36

Internal/support tool for locating the JS source of one concrete encrypted request field, sign, x-sign, authKey, or token. Use directly only when the user explicitly asks for an atomic entry-location task such as "找加密入口", "签名怎么算", or "这个请求头字段哪里生成"; otherwise let reverse-js-crawler or website-314-api-delivery choose it. Do not trigger for generic crawler delivery, ordinary request inspection, challenge/WAF backend acceptance, or non-encrypted parameters.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

출처

zy950618

zy950618/oh_my_reverse_skill

GitHub 저장소 열기 Creator 저장소 보기

다운로드

find-crypto-entry

Hardened Tool Governance

Version: 0.5.0

Change log: 0.5.0 adds structured eval coverage, graph/impact examples, and a hard delivery gate for evidence-first crypto-entry work.

Workflow:

Use references/governance.md before final output, especially when the target field affects endpoints, request headers, request body, cache state, session state, or evals.
Use references/graph-impact-examples.md when writing node/edge updates or impact-regression records.
Treat the final observed API request/response and captured call stack as the source of truth. Do not infer hidden endpoints, parameters, cookies, tokens, device IDs, signatures, or risk tokens.

Success Criteria:

Output the parameter, script URL, line/column or searchable offset, function name, call path, observed request URL, and evidence pointer.
Classify every important claim as observed, derived, assumed, or unverified.
Update 站点经验库/<domain>/knowledge-graph.md and 站点经验库/<domain>/impact-regression.md when the entry point, endpoint, field, protection state, or eval changes.
List commands/tools used and the required replay/diff/schema regression checks.

Evidence Discipline:

Algorithm restoration, request replay, and runtime environment work require their own evidence and downstream skill handoff when they go beyond entry location.
Fingerprint and risk-control fields must carry source, capture id, replay effect, and freshness status before reuse.
Do not hardcode request headers, fingerprints, cookies, tokens, signatures, timestamps, or session-derived values.

Governance:

Before expanding across market/stage/session, require separate evidence for each scope.
Before claiming concurrency support, require a concurrency ladder and session/cache/fingerprint isolation evidence.
If evidence is missing, stop at unverified and write the missing capture requirement instead of guessing.

定位加密参数 $ARGUMENTS 的生成入口。

目标：找到生成该参数的函数位置（脚本 URL + 行列号 + 函数名 + 调用路径）。

策略选择

根据信号选择策略，不要按固定顺序执行：

优先静态搜索：用 search_in_sources 搜参数名（带 excludeMinified=false）。大多数情况下直接找到赋值位置，然后读上下文追溯来源。

静态搜不到时用 XHR 断点：break_on_xhr 设在请求 URL 特征上，刷新页面触发。断点命中后从调用栈中找业务代码帧，用 evaluate_on_callframe 检查变量。

两种策略可以组合：先静态搜索定位赋值位置，再在赋值处设断点动态验证。

领域知识

agent 在逆向分析中容易踩的坑，这些是无法通过推理得出的经验：

加密参数的 3 种常见架构

业务代码直接赋值 — 在请求函数中 headers["x-sign"] = encrypt(data)。静态搜索直接找到。
请求拦截器统一加签 — axios interceptor 或 fetch wrapper 中统一添加。搜参数名可能只在拦截器中出现一次。
外部安全 SDK — 独立 JS 文件（通常混淆）挂载全局对象（如 window.h5sign），业务代码调用其方法。静态搜索能找到调用处，但 SDK 内部代码全被混淆。

OB 混淆的影响

当加密逻辑在 OB 混淆的文件中（特征：_0x 前缀、大型字符串数组、RC4 解密函数），所有字符串都被加密，静态搜索在该文件内无法匹配任何明文。但调用该文件的业务代码通常未混淆，从业务代码侧搜索更高效。

XHR 断点命中时的堆栈特点

XHR 断点命中在 send() 调用处，调用栈底部通常是框架代码（axios/fetch wrapper）。加密逻辑在栈的中上层。直接跳过底部框架帧，关注业务代码帧。

反模式

不要反复 step_into — 容易掉进框架响应式系统（Vue reactivity、React fiber）。用 evaluate_on_callframe 直接检查变量比单步跟踪高效得多。
不要在 OB 混淆文件中搜字符串 — 所有明文都被加密了，搜不到。从非混淆的调用方搜。
不要频繁刷新页面 — 每次刷新所有 scriptId 失效。设好断点再刷新，一次到位。
断点命中时优先用 evaluate_on_callframe 指定帧 — evaluate_script 在暂停时会自动 fallback 到顶层帧，但如果需要检查特定调用帧的变量，必须用 evaluate_on_callframe 指定 frameIndex。

Tool Policy

开始实现前 Read ~/.claude/skills/karpathy-guidelines/SKILL.md,确认 4 条原则:Think Before Coding / Simplicity First / Surgical Changes / Goal-Driven Execution。这是基础层规范,所有执行类 skill 强制依赖。
遇到逆向运行时问题(断点/时间/cookie/TLS 指纹/风控恢复/接口变更)Read ~/.claude/skills/oh_my_reverse_skill/99-SKILLS治理/10-逆向运行时常见问题.md。
输出结论、扩范围或做并发前 Read ~/.claude/skills/oh_my_reverse_skill/99-SKILLS治理/11-AI事实证据规约.md / 12-反泛化与任务收敛规约.md / 13-并发指纹与会话隔离规约.md。
改端点/字段/状态/保护/实现/eval 前后 Read 14-知识图谱行程与关联规约.md / 15-AI变更风险与回归校验规约.md,并更新 knowledge-graph.md / impact-regression.md。

完成标准

找到以下信息即为完成：

入口位置：
- 参数：$ARGUMENTS
- 脚本：https://example.com/static/js/main.abc123.js
- 位置：第 X 行，第 Y 列
- 函数：functionName (或 anonymous)
- 调用路径：request → addSign → encrypt
- 加密类型：[标准算法名 | 外部SDK | 未知/需解混淆]

只找入口，不做算法还原。

이 저장소의 다른 Skills

같은 저장소

skills-evaluation-governance

zy950618/oh_my_reverse_skill

Use this skill to score, refine, backtest, and govern Codex/Claude skills: assess whether notes are installable skills, create evals, prepare Skill Bench structure, compare SKILL.md quality, define trigger/negative-trigger tests, track drift, maintain versions, record changes, and enforce admission gates for new Skills. Trigger when the user asks to rate skills, use Skill Bench, convert notes into usable skills, evaluate trigger accuracy, maintain a skills library, score a new Skill before accepting it, run backtests, or Chinese requests such as SKILLS评分, 技能评分, 可用SKILLS, 新增Skill准入, Skill Bench跑分, 回测, 漂移测试, 长期治理, 版本号, 变更记录, 触发词优化, 负例测试, or 技能库治理.

2026-07-1210

web-h5-loop-engineering

zy950618/oh_my_reverse_skill

Orchestration entry for Web/H5 reverse-engineering work only when the user explicitly asks for LOOP, closed-loop handling, multi-agent/three-role verification, repeated validation, execution ledger, acceptance report, or when a prior attempt failed because evidence, repeat verification, cleanup, impact, or backend acceptance was incomplete. It coordinates executor, verifier, and governor roles with loop ledger, acceptance report, fixture freshness, and metrics. Do not trigger for ordinary one-pass crawler tasks, simple fixture freshness checks, or single-tool JS work; use reverse-js-crawler or the relevant support tool first.

2026-07-1210

env-patch

zy950618/oh_my_reverse_skill

Internal/support tool for running a previously identified browser encryption module in Node.js with env_core.js, prototype/native-function shims, and minimal stubs. Use directly only when the user explicitly asks for 补环境, Node里跑, webpack模块提取, or when an entry skill has already located the module and needs browser-to-Node reproduction. Do not trigger for ordinary browser debugging, AST deobfuscation, generic Node.js code, crawler delivery, or unresolved API discovery.

2026-07-0910

js-page-runtime-parity

zy950618/oh_my_reverse_skill

Extract page-level JavaScript runtime dependencies and verify Browser, Node, V8, and PageRuntime output parity for authorized targets or localhost labs without generating risk tokens or defeat behavior.

2026-07-0910

authorized-target-adapter

zy950618/oh_my_reverse_skill

Adapt reverse-engineering workflows only for owned or explicitly authorized targets with scope, allowed hosts, rate limits, stop conditions, redaction, and business data assertions.

2026-07-0810

reverse-js-crawler

zy950618/oh_my_reverse_skill

Primary entry for focused Web/H5 crawler reverse engineering and interface restoration: page reconnaissance, real API discovery, JavaScript sign/token/cookie source tracing, request reproduction, data collection scripts, UI/API parity, and fresh replay evidence for a bounded route or stage. Trigger for JS reverse, crawler reverse, API restoration, encrypted parameters, sign/x-sign/authKey source tracing, web data collection, request reproduction, 逆向采集, JS逆向, 接口还原, 接口复现, 加密参数, 请求复现, 批量采集, 数据清洗, or 采集脚本交付. Do not use as the first entry for full FastAPI/service delivery, explicit LOOP/multi-agent closure, skill governance, stable adapter production, or challenge/WAF-specific evidence; route those to the corresponding entry or conditional skill.

2026-07-0810

name	find-crypto-entry
standard_type	internal_tool
description	Internal/support tool for locating the JS source of one concrete encrypted request field, sign, x-sign, authKey, or token. Use directly only when the user explicitly asks for an atomic entry-location task such as "找加密入口", "签名怎么算", or "这个请求头字段哪里生成"; otherwise let reverse-js-crawler or website-314-api-delivery choose it. Do not trigger for generic crawler delivery, ordinary request inspection, challenge/WAF backend acceptance, or non-encrypted parameters.
argument-hint	["参数名"]
platforms	["web","h5"]

find-crypto-entry

Hardened Tool Governance

Version: 0.5.0

Change log: 0.5.0 adds structured eval coverage, graph/impact examples, and a hard delivery gate for evidence-first crypto-entry work.

Workflow:

Use references/governance.md before final output, especially when the target field affects endpoints, request headers, request body, cache state, session state, or evals.
Use references/graph-impact-examples.md when writing node/edge updates or impact-regression records.
Treat the final observed API request/response and captured call stack as the source of truth. Do not infer hidden endpoints, parameters, cookies, tokens, device IDs, signatures, or risk tokens.

Success Criteria:

Output the parameter, script URL, line/column or searchable offset, function name, call path, observed request URL, and evidence pointer.
Classify every important claim as observed, derived, assumed, or unverified.
Update 站点经验库/<domain>/knowledge-graph.md and 站点经验库/<domain>/impact-regression.md when the entry point, endpoint, field, protection state, or eval changes.
List commands/tools used and the required replay/diff/schema regression checks.

Evidence Discipline:

Algorithm restoration, request replay, and runtime environment work require their own evidence and downstream skill handoff when they go beyond entry location.
Fingerprint and risk-control fields must carry source, capture id, replay effect, and freshness status before reuse.
Do not hardcode request headers, fingerprints, cookies, tokens, signatures, timestamps, or session-derived values.

Governance:

Before expanding across market/stage/session, require separate evidence for each scope.
Before claiming concurrency support, require a concurrency ladder and session/cache/fingerprint isolation evidence.
If evidence is missing, stop at unverified and write the missing capture requirement instead of guessing.

定位加密参数 $ARGUMENTS 的生成入口。

目标：找到生成该参数的函数位置（脚本 URL + 行列号 + 函数名 + 调用路径）。

策略选择

根据信号选择策略，不要按固定顺序执行：

优先静态搜索：用 search_in_sources 搜参数名（带 excludeMinified=false）。大多数情况下直接找到赋值位置，然后读上下文追溯来源。

静态搜不到时用 XHR 断点：break_on_xhr 设在请求 URL 特征上，刷新页面触发。断点命中后从调用栈中找业务代码帧，用 evaluate_on_callframe 检查变量。

两种策略可以组合：先静态搜索定位赋值位置，再在赋值处设断点动态验证。

领域知识

agent 在逆向分析中容易踩的坑，这些是无法通过推理得出的经验：

加密参数的 3 种常见架构

业务代码直接赋值 — 在请求函数中 headers["x-sign"] = encrypt(data)。静态搜索直接找到。
请求拦截器统一加签 — axios interceptor 或 fetch wrapper 中统一添加。搜参数名可能只在拦截器中出现一次。
外部安全 SDK — 独立 JS 文件（通常混淆）挂载全局对象（如 window.h5sign），业务代码调用其方法。静态搜索能找到调用处，但 SDK 内部代码全被混淆。

OB 混淆的影响

XHR 断点命中时的堆栈特点

XHR 断点命中在 send() 调用处，调用栈底部通常是框架代码（axios/fetch wrapper）。加密逻辑在栈的中上层。直接跳过底部框架帧，关注业务代码帧。

反模式

不要反复 step_into — 容易掉进框架响应式系统（Vue reactivity、React fiber）。用 evaluate_on_callframe 直接检查变量比单步跟踪高效得多。
不要在 OB 混淆文件中搜字符串 — 所有明文都被加密了，搜不到。从非混淆的调用方搜。
不要频繁刷新页面 — 每次刷新所有 scriptId 失效。设好断点再刷新，一次到位。
断点命中时优先用 evaluate_on_callframe 指定帧 — evaluate_script 在暂停时会自动 fallback 到顶层帧，但如果需要检查特定调用帧的变量，必须用 evaluate_on_callframe 指定 frameIndex。

Tool Policy

开始实现前 Read ~/.claude/skills/karpathy-guidelines/SKILL.md,确认 4 条原则:Think Before Coding / Simplicity First / Surgical Changes / Goal-Driven Execution。这是基础层规范,所有执行类 skill 强制依赖。
遇到逆向运行时问题(断点/时间/cookie/TLS 指纹/风控恢复/接口变更)Read ~/.claude/skills/oh_my_reverse_skill/99-SKILLS治理/10-逆向运行时常见问题.md。
输出结论、扩范围或做并发前 Read ~/.claude/skills/oh_my_reverse_skill/99-SKILLS治理/11-AI事实证据规约.md / 12-反泛化与任务收敛规约.md / 13-并发指纹与会话隔离规约.md。
改端点/字段/状态/保护/实现/eval 前后 Read 14-知识图谱行程与关联规约.md / 15-AI变更风险与回归校验规约.md,并更新 knowledge-graph.md / impact-regression.md。

完成标准

找到以下信息即为完成：

入口位置：
- 参数：$ARGUMENTS
- 脚本：https://example.com/static/js/main.abc123.js
- 位置：第 X 行，第 Y 列
- 函数：functionName (或 anonymous)
- 调用路径：request → addSign → encrypt
- 加密类型：[标准算法名 | 外部SDK | 未知/需解混淆]

只找入口，不做算法还原。