인터페이스는 현지화되어 있으며, Skill 소개와 설명 문서는 선택한 언어로 AI가 자동 번역합니다. 코드·명령어·경로·전문 용어는 원문을 유지하며 원작자 문서가 우선합니다.
Skills 둘러보기
생산성 및 협업 · SKILL.md

poam-automation-expert

poam-automation-expert

Expertise in FedRAMP POA&M lifecycle management, FedRAMP 20x VDR generation, and vulnerability classification using CISA KEV, EPSS, N-ratings, LEV/IRV, and NIST 800-53 control mappings.

Independent GitHub snapshot · Automated classification · Original documentation governs

Skill 문서

GitHub

name: poam-automation-expert description: "Expertise in FedRAMP POA&M lifecycle management, FedRAMP 20x VDR generation, and vulnerability classification using CISA KEV, EPSS, N-ratings, LEV/IRV, and NIST 800-53 control mappings."

POA&M Automation Expert

Tool Behavior

This connector wraps the FedRAMP POA&M Automation Tool — a Python CLI that manages the full vulnerability finding lifecycle from scanner import through FedRAMP 20x VDR generation.

What it does

  • Parses raw scanner exports (Nessus, Tenable, Qualys, Wiz) into a structured FedRAMP POA&M
  • Auto-deduplicates findings across scanner imports using title + asset identifier
  • Enriches findings with NIST 800-53 control mappings via keyword matching
  • Manages finding lifecycle: update, close, deviation (false positive / operational requirement)
  • Generates FedRAMP 20x VDR reports with live CISA KEV and EPSS enrichment
  • Generates per-release Product VDR reports for customer-facing vulnerability disclosure
  • Outputs machine-readable JSON alongside every Excel report

VDR Classification Logic

Each finding is classified using live threat intelligence:

  • CISA KEV: Finding CVE is checked against the live CISA Known Exploited Vulnerabilities catalog. KEV findings are bumped to minimum N3 adverse impact.
  • EPSS: Exploit Prediction Scoring System score fetched from First.org API. Score ≥ 0.10 triggers LEV classification.
  • LEV/NLEV: Likely Exploitable / Not Likely Exploitable. Determined by KEV membership, EPSS score, and severity fallback.
  • IRV/NIRV: Internet Reachable / Not Internet Reachable. Derived from keywords in asset identifier and weakness description.
  • N-Rating: N1 (lowest) to N5 (highest) adverse impact. Base rating derived from severity, then adjusted upward for KEV and IRV+LEV combinations.

Evaluation Deadlines

Per FedRAMP 20x baseline:

  • High: 2 days from detection
  • Moderate: 5 days from detection
  • Low: 7 days from detection

Auto-Acceptance Threshold

Findings open 192+ days are automatically moved to the Accepted Vulnerabilities sheet per VDR-TFR-MAV.

Update Frequency Requirements

  • High baseline: Weekly (every 7 days)
  • Moderate baseline: Every 14 days
  • Low baseline: Monthly (every 30 days)

Control Coverage

This connector maps findings to NIST 800-53 Rev 5 controls via keyword matching across 30+ control families including SI-2, RA-5, AC-2, AC-3, SC-8, SC-28, AU-2, CM-6, CM-7, and others.

Failure Modes

  • Missing openpyxl: Install with pip install openpyxl
  • CISA KEV unavailable: Offline fallback uses severity-based LEV classification
  • EPSS unavailable: Offline fallback uses severity-based LEV classification
  • POA&M file open in Excel: Close the file before running any write command
  • Duplicate findings: Automatically skipped on import — same title + asset identifier

Limitations

  • Product VDR requires a manually maintained vulnerability CSV — it does not auto-discover product vulnerabilities
  • NIST control mapping is keyword-based, not authoritative — review mappings before submission
  • AI remediation text requires Ollama running locally with the Mistral model
저장한 Skills