cve-poc-generator — Claude Code skill: what it does, how to install
The cve-poc-generator skill researches a CVE, writes a safe Python PoC, and produces a vulnerability report. Use it when you need sourced verification and remediation details.
MIT 236 installs ★ 632 updated 2026-07-29
npx skills add transilienceai/communitytools --skill cve-poc-generator -a claude-codeWhat cve-poc-generator does
cve-poc-generator takes one CVE ID and pulls records from the NVD API v2.0. It also examines vendor advisories, GitHub Security Advisories, Exploit-DB, and published analyses. The workflow extracts the applicable CVSS score and vector, CWE IDs, affected product matches, advisory references, and patch links before documenting the root cause and attack path.
The generated files are poc.py under artifacts/cve-pocs/CVE-XXXX-XXXXX/ and report.md under reports/cve-pocs/CVE-XXXX-XXXXX/. The PoC is meant to run with Python’s standard library plus requests, without framework dependencies. Its default behavior is read-only and non-destructive; any risky operation needs an explicit --confirm flag. Reports attribute claims to sources, and CVSS values come from NVD rather than estimation.
What it helps with
- research a CVE from its identifier
- collect CVSS, CWE, and affected versions
- write a standalone Python verification PoC
- produce a sourced report with remediation guidance
Example prompts for the agent
- Run
/cve-poc-generator CVE-2024-XXXXXand create the PoC with its report. - Research
/cve-poc-generator CVE-2024-XXXXX, check NVD references, and save the findings toreport.md. - Process
/cve-poc-generator CVE-2024-XXXXXusing read-only checks and no destructive payloads.
How to install cve-poc-generator
1Skills CLI (Vercel)
npx skills add transilienceai/communitytools --skill cve-poc-generator -a claude-code2Manually: put the files in place
~/.claude/skills/cve-poc-generator/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/cve-poc-generator/
1Skills CLI (Vercel)
npx skills add transilienceai/communitytools --skill cve-poc-generator -a codex2Manually: put the files in place
.agents/skills/cve-poc-generator/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add transilienceai/communitytools --skill cve-poc-generator -a cursor2Manually: put the files in place
.cursor/skills/cve-poc-generator/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install transilienceai/communitytools skills/cve-poc-generator2Manually: put the files in place
.github/skills/cve-poc-generator/SKILL.md1Skills CLI (Vercel)
npx skills add transilienceai/communitytools --skill cve-poc-generator -a gemini-cli2Platform CLI
gemini skills install https://github.com/transilienceai/communitytools --path skills/cve-poc-generatorManually: put the files in place
.devin/skills/cve-poc-generator/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add transilienceai/communitytools --skill cve-poc-generator -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about cve-poc-generator
How do I install cve-poc-generator in Claude Code?
Run npx skills add transilienceai/communitytools --skill cve-poc-generator -a claude-code, then use ~/.claude/skills/cve-poc-generator/SKILL.md.
What does cve-poc-generator do?
It researches a CVE through NVD and advisory sources, creates a standalone Python PoC, and writes a Markdown report covering the cause, risk, evidence, and remediation.
Can one run process multiple CVEs?
No. Each invocation accepts one CVE ID, so multiple CVEs require separate invocations.
Source text
SKILL.md · MIT
---
name: cve-poc-generator
description: CVE research, standalone PoC script and report generation. Given a CVE ID, researches NVD and advisories, generates a safe Python PoC, and writes a detailed vulnerability report.
---
# CVE PoC Generator
Research a CVE by ID, generate a standalone Python proof-of-concept script, and produce a detailed vulnerability report.
## Workflow
1. **NVD Lookup** - Query NVD API v2.0 for the CVE ID (`python3 tools/nvd-lookup.py <CVE>`). Extract the CVSS score/vector using the v4.0-primary ladder (v4.0 → v3.1 → v3.0 → v2.0; the tool's `JSON_SUMMARY` exposes `cvss_version` + `cvss_vector`), plus CWE IDs, CPE matches, advisory URLs, and patch links.
2. **Advisory Research** - Deep-dive vendor advisories, GitHub security advisories, Exploit-DB, and published write-ups. Identify root cause, affected versions, and attack vector details.
3. **PoC Generation** - Write a standalone Python script (`poc.py`) that demonstrates the vulnerability safely. Follow the script standards in `reference/poc-methodology.md`.
4. **Report Generation** - Write a comprehensive markdown report (`report.md`) with metadata, root cause analysis, risk assessment, and remediation guidance.
## NVD Data to Collect
| Field | Source | Usage |
|-------|--------|-------|
| CVE ID | NVD | Primary identifier |
| CVSS Score + Vector (v4.0 primary; v3.1/v3.0/v2.0 fallback) | NVD | Risk scoring |
| CWE ID(s) | NVD | Vulnerability classification |
| CPE Matches | NVD | Affected products and versions |
| Advisory URLs | NVD references | Research sources |
| Patch Links | NVD references / vendor | Remediation guidance |
| Description | NVD | Vulnerability summary |
| Published / Modified dates | NVD | Timeline |
## Output
```
{OUTPUT_DIR}/
artifacts/cve-pocs/CVE-XXXX-XXXXX/
poc.py # Standalone Python PoC script
reports/cve-pocs/CVE-XXXX-XXXXX/
report.md # Detailed vulnerability report
```
## Invocation
```
/cve-poc-generator CVE-2024-XXXXX
```
The skill accepts a single CVE ID as argument. Multiple CVEs should be processed with separate invocations.
## Rules
1. **Least harm** - PoC scripts MUST demonstrate vulnerability without causing damage. Use detection/verification checks, not destructive payloads.
2. **Standalone scripts** - PoC must run independently with only standard Python libraries plus `requests`. No framework dependencies.
3. **Accurate scoring** - Use the exact CVSS score and vector from NVD. Do not fabricate or estimate scores.
4. **Source attribution** - Every claim in the report must cite its source (NVD, vendor advisory, CVE description).
5. **No emoji** - Use text severity labels only (CRITICAL, HIGH, MEDIUM, LOW, INFORMATIONAL).
6. **Verified data only** - Do not hallucinate CVE details. If NVD data is unavailable, state it explicitly.
7. **Safe defaults** - PoC scripts must default to read-only, non-destructive operations. Any potentially harmful action requires explicit `--confirm` flag.
© transilienceai · MIT · transilienceai/communitytools