validator — Claude Code skill: vet findings before reporting
validator — a skill for vetting bug bounty findings before writing the report. Runs a 7-question system and 4-gate checklist, filters weak and theoretical bugs in minutes, and outputs a decision: PASS, KILL, DOWNGRADE, or CHAIN REQUIRED.
no license ★ 1,298 updated 2026-06-12
npx skills add H-mmer/pentest-agents --skill validator -a claude-codeWhat validator does
validator works across Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI, and Windsurf. It catches mistakes that waste hours: treating a 200 status as proof, testing IDOR on one account instead of two, reporting CORS without credentialed exfil, self-XSS, or standalone open redirects. Give it a finding description; get a clear decision with a single reason to reject or a green light for the report.
The skill runs 7 questions in order: is there a real HTTP request, is the bug in scope, do you need admin rights, is it already known, can you prove impact beyond the code. Then Gate 2 checks for duplicates across HackerOne Hacktivity and changelogs. No hand-waving about possible chains — either the finding works now or you build the specific chain it needs.
Requires Python and access to H-mmer/pentest-agents. Built for bug bounty and pentesting experience; without it, the checklist demands extra reading, but remains a tool, not a guide.
What it helps with
- filter weak findings before writing the report
- decide whether a bug needs a chain to be valid
- check if the bug fits the program scope
- tell a real proof-of-concept from theoretical impact
- spot duplicates in HackerOne Hacktivity in seconds
Example prompts for the agent
- Vet a finding: SQL injection in
POST /api/searchquery parameter. Do you have the exact request and response showing the leak? - Validate IDOR on
GET /users/{id}/profile— did another account's data return when you switched accounts? - Run validator on an open redirect to learn if it needs an OAuth token theft chain or gets killed standalone.
How to install validator
1Skills CLI (Vercel)
npx skills add H-mmer/pentest-agents --skill validator -a claude-code2Manually: put the files in place
~/.claude/skills/agent-validator/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/agent-validator/
1Skills CLI (Vercel)
npx skills add H-mmer/pentest-agents --skill validator -a codex2Manually: put the files in place
.agents/skills/agent-validator/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add H-mmer/pentest-agents --skill validator -a cursor2Manually: put the files in place
.cursor/skills/agent-validator/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install H-mmer/pentest-agents providers/cursor/.cursor/skills/agent-validator2Manually: put the files in place
.github/skills/agent-validator/SKILL.md1Skills CLI (Vercel)
npx skills add H-mmer/pentest-agents --skill validator -a gemini-cli2Platform CLI
gemini skills install https://github.com/H-mmer/pentest-agents --path providers/cursor/.cursor/skills/agent-validatorManually: put the files in place
.devin/skills/agent-validator/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.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 validator
How do I install validator?
It depends on your IDE. For Claude Code: npx skills add H-mmer/pentest-agents --skill agent-validator -a claude-code, then open ~/.claude/skills/agent-validator/SKILL.md. For Cursor: npx skills add H-mmer/pentest-agents --skill agent-validator -a cursor and .cursor/skills/agent-validator/SKILL.md. For GitHub Copilot: gh skill install H-mmer/pentest-agents agent-validator.
What does validator do?
It runs a finding through 7 questions (real request, in-scope, needs admin rights) and a 4-gate check (impact, dedup, report quality). Outputs one decision: PASS, KILL with the question number, DOWNGRADE, or CHAIN REQUIRED with the exact chain needed.
What bugs does validator catch most often?
Status 200 with no real exploit, IDOR on a single account instead of cross-account, CORS wildcard without credentialed exfil, theoretical language (could lead to) instead of real HTTP proof, findings from the never-submit list without a valid chain.