菜单
Apply bottom-up and top-down role mining techniques to discover optimal RBAC roles from existing user-permission assignments, reducing role explosion and enforcing least privilege.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/xalgord/xalgorix --skill building-role-mining-for-rbac-optimization命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Detecting and exploiting HTTP request smuggling vulnerabilities caused by Content-Length and Transfer-Encoding parsing discrepancies between front-end and back-end servers.
Identifying and exploiting OAuth 2.0 and OpenID Connect misconfigurations including redirect URI manipulation, token leakage, and authorization code theft during security assessments.
Identifying and exploiting dangling DNS records pointing to unclaimed cloud services, enabling subdomain takeover for phishing, cookie stealing, and authentication bypass during authorized penetration tests.
基于 SOC 职业分类
正在显示 SKILL.md
| name | building-role-mining-for-rbac-optimization |
| description | Apply bottom-up and top-down role mining techniques to discover optimal RBAC roles from existing user-permission assignments, reducing role explosion and enforcing least privilege. |
| domain | cybersecurity |
| subdomain | identity-access-management |
| tags | ["rbac","role-mining","identity-governance","access-control","least-privilege","clustering"] |
| version | 1.0 |
| author | mahipal |
| license | Apache-2.0 |
| nist_csf | ["PR.AA-01","PR.AA-02","PR.AA-05","PR.AA-06"] |
Role mining is the process of analyzing existing user-permission assignments to discover optimal roles for a Role-Based Access Control (RBAC) system. Organizations accumulate excessive permissions over time through job changes, project assignments, and ad-hoc access grants, leading to "role explosion" where thousands of granular roles exist with significant overlap. Role mining uses data analysis -- including clustering algorithms, formal concept analysis, and graph-based methods -- to consolidate permissions into a minimal set of roles that accurately represent business functions while enforcing least privilege.
core_permissions threshold down (e.g. >50% instead of >80%) inflates coverage to "95%" by handing every cluster member permissions only half of them actually use. Verify deviation/extra-permission rate stays <5% alongside the coverage figure — high coverage with high deviation means the role grants more than people hold.R × P and diff against the original UPA matrix — every removed permission must be an intended least-privilege reduction, every added permission flagged; re-run evaluate_role_set and assert coverage >95% AND deviation <5% together.| Approach | Description | Best For |
|---|---|---|
| Bottom-Up | Analyze existing permissions to discover common patterns | Large datasets with organic permission growth |
| Top-Down | Design roles from business requirements and job descriptions | Greenfield RBAC or organizational restructuring |
| Hybrid | Combine bottom-up analysis with top-down business validation | Most production environments |
1. Permission Clustering: Group users with similar permission sets using k-means or hierarchical clustering. Users in the same cluster share a common role.
2. Formal Concept Analysis (FCA): Mathematical framework that identifies complete set of concepts (user groups sharing exact permission sets) from a binary user-permission matrix.
3. Graph-Based Mining: Model users and permissions as a bipartite graph, then find dense subgraphs representing candidate roles.
4. Boolean Matrix Decomposition: Decompose the user-permission matrix U into U ≈ R × P where R maps users to roles and P maps roles to permissions.
| Metric | Formula | Target |
|---|---|---|
| Role Count | Total distinct roles after mining | Minimize |
| Coverage | Permissions explained by mined roles / Total permissions | > 95% |
| Weighted Structural Complexity (WSC) | Sum of role-user + role-permission assignments | Minimize |
| Deviation | Extra permissions not covered by assigned roles | < 5% |
Collect the current access state from all identity sources:
import pandas as pd
import numpy as np
# Load user-permission assignments
# Format: user_id, permission_id (one row per assignment)
assignments = pd.read_csv("user_permissions.csv")
# Create binary user-permission matrix (UPA matrix)
upa_matrix = assignments.pivot_table(
index="user_id",
columns="permission_id",
aggfunc="size",
fill_value=0
)
upa_matrix = (upa_matrix > 0).astype(int)
print(f"Users: {upa_matrix.shape[0]}")
print(f"Permissions: {upa_matrix.shape[1]}")
print(f"Assignments: {assignments.shape[0]}")
print(f"Density: {upa_matrix.values.sum() / upa_matrix.size:.2%}")
from sklearn.cluster import AgglomerativeClustering
from sklearn.metrics import silhouette_score
def find_optimal_clusters(matrix, max_k=50):
"""Find optimal number of roles using silhouette analysis."""
scores = []
for k in range(2, min(max_k, matrix.shape[0])):
clustering = AgglomerativeClustering(
n_clusters=k, metric="jaccard", linkage="average"
)
labels = clustering.fit_predict(matrix)
score = silhouette_score(matrix, labels, metric="jaccard")
scores.append((k, score))
optimal_k = max(scores, key=lambda x: x[1])[0]
return optimal_k, scores
def mine_roles_clustering(upa_matrix, n_clusters):
"""Mine roles using hierarchical clustering on Jaccard distance."""
clustering = AgglomerativeClustering(
n_clusters=n_clusters, metric="jaccard", linkage="average"
)
user_matrix = upa_matrix.values
labels = clustering.fit_predict(user_matrix)
roles = {}
for cluster_id in range(n_clusters):
cluster_users = upa_matrix.index[labels == cluster_id]
cluster_permissions = upa_matrix.loc[cluster_users]
# Core role = permissions held by >80% of cluster members
permission_frequency = cluster_permissions.mean()
core_permissions = permission_frequency[permission_frequency >= 0.8].index.tolist()
roles[f"Role_{cluster_id}"] = {
"permissions": core_permissions,
"user_count": len(cluster_users),
"users": cluster_users.tolist(),
"coverage": permission_frequency[permission_frequency >= 0.8].mean()
}
return roles, labels
def mine_roles_fca(upa_matrix, min_support=3):
"""Mine roles using Formal Concept Analysis (frequent closed itemsets)."""
from itertools import combinations
users = upa_matrix.index.tolist()
permissions = upa_matrix.columns.tolist()
concepts = []
# Find all maximal permission sets shared by at least min_support users
for size in range(len(permissions), 0, -1):
for perm_combo in combinations(permissions, size):
perm_set = set(perm_combo)
# Find users who have ALL permissions in this set
matching_users = []
for user in users:
user_perms = set(upa_matrix.columns[upa_matrix.loc[user] == 1])
if perm_set.issubset(user_perms):
matching_users.append(user)
if len(matching_users) >= min_support:
# Check if this is a closed concept (no superset with same extent)
is_closed = True
for concept in concepts:
if set(matching_users) == set(concept["users"]) and \
perm_set.issubset(set(concept["permissions"])):
is_closed = False
break
if is_closed:
concepts.append({
"permissions": list(perm_set),
"users": matching_users,
"support": len(matching_users)
})
if len(concepts) > 100: # Limit for performance
break
return concepts
def evaluate_role_set(roles, upa_matrix):
"""Evaluate the quality of a mined role set."""
total_assignments = upa_matrix.values.sum()
covered_assignments = 0
extra_assignments = 0
for role_name, role_data in roles.items():
role_perms = set(role_data["permissions"])
for user in role_data["users"]:
user_perms = set(upa_matrix.columns[upa_matrix.loc[user] == 1])
covered = role_perms.intersection(user_perms)
extra = role_perms - user_perms
covered_assignments += len(covered)
extra_assignments += len(extra)
metrics = {
"total_roles": len(roles),
"total_assignments": total_assignments,
"covered_assignments": covered_assignments,
"coverage_rate": covered_assignments / total_assignments if total_assignments else 0,
"extra_permissions": extra_assignments,
"deviation_rate": extra_assignments / (covered_assignments + extra_assignments) if (covered_assignments + extra_assignments) else 0,
"avg_role_size": np.mean([len(r["permissions"]) for r in roles.values()]),
"avg_users_per_role": np.mean([r["user_count"] for r in roles.values()]),
}
return metrics
After mining candidate roles: