validator · h-mmer — Claude Code subagent: filter weak findings fast
validator — a triage subagent for security researchers that kills weak findings before report writing using a 7-question gate and 4-gate validation checklist. Runs real HTTP proofs, not code review.
no license ★ 1,298 updated 2026-06-12
.claude/agents/validator.mdWhat validator does
validator asks each finding seven questions in sequence: can a real HTTP request exploit it, does it fit the program scope, does it require privilege an attacker cannot obtain, is it already known, is impact proved beyond theory, and is it on the never-submit list. A single no answer ends the finding.
Then come four gates: confirmation via real requests and scope check, dedup search through the writeups database, report quality review, and CVSS calculation. Output is one of four: PASS, KILL with the failed question number, DOWNGRADE if severity is overclaimed, or CHAIN REQUIRED if the finding chains to another but fails on its own.
The subagent uses bash, file read, grep, and MCP integration with a writeups search server to check prior disclosures. It requires actual server responses, not code paths or hypothetical impact. Disallowed tools: Write and Edit, keeping it read-only during triage.
What it helps with
- filter theoretical vulnerabilities before writing a report
- reject a finding that was already disclosed or is on the program blacklist
- prove a vulnerability works with a real HTTP request and server response, not code analysis
- verify scope and confirm the finding does not require administrator privileges
Example prompts for the agent
- Feed finding details to validator and it runs the 7-question gate then the 4 validation gates
- Ensure you have the real HTTP request and server response attached; validator kills on Q1 if they are missing
- If a finding chains to another but fails alone, validator returns CHAIN REQUIRED with the specific chain needed
How to install validator
Manually: put the files in place
.claude/agents/validator.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/validator.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/validator.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about validator
How do I install validator in Claude Code?
Copy the file to .claude/agents/validator.md in your project.
What does validator do?
It triages security findings through a 7-question gate and 4-gate validation, rejecting weak or theoretical vulnerabilities before report submission.
What does CHAIN REQUIRED mean?
The finding fails on its own but becomes valid when chained with another vulnerability; validator specifies which chain to build.
Similar subagents
All subagents for Security →validator by other authors comes first.