Numerical QA Agent — Testing & QA agent for Claude Code
Independently recomputes every displayed number and writes fail-closed numerical QA evidence.
How to install Numerical QA Agent
Installs to ~/.claude/agents/mlobo2012-round-finance-agents-numerical-qa-agent.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/mlobo2012/round-finance-agents/HEAD/agents/numerical-qa-agent.md -o ~/.claude/agents/mlobo2012-round-finance-agents-numerical-qa-agent.md Restart Claude Code, or start a new session, for it to be picked up.
What Numerical QA Agent does
name: numerical-qa-agent description: Independently recomputes every displayed number and writes fail-closed numerical QA evidence. tools: Read, Grep, Glob
Numerical QA Agent
You are the independent numerical reviewer. Your job is to fail closed.
Inputs
contract.jsonsource_ledger.jsongenerator_output.mdor extracted PPTX content- extracts and workpapers
- Round MCP evidence summaries supplied by the main session, including Xero data returned through Round MCP when
Alternatives in Testing & QA
- Civitai Test Review — Reviews the tests in a feature segment of the main Civitai Next.js app (src/) for whether they would actually 7.2k ★
- Fable Verifier — Cold verifier for finished fable deliverables 837 ★
- Witness — Independently witness that an Allium loop's convergence claim is true and was reached honestly 474 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Math Reviewer
Independent verifier for mathematical and numerical work. Mode "spec" recomputes and checks a math-spec deriva
Generator Agent
Drafts CFO artifacts while tagging every displayed number for independent QA.
Claude Grader
Grades one extension test run against its assertions and writes grading.json. Use after an eval run finishes,
Cha Mutant Smith
Writes mutants that break one meaning-carrying line each, with anchors lifted verbatim and a real test path. U
Spec Verifier
Spec compliance verifier - independently checks the implementation against the plan's acceptance criteria and
P2a Visual Reviewer
Independently compares rendered UI evidence with the approved Plan2Agent experience spec and HTML prototype, g
Related Skills
Docforge
Writes new repository documentation from source-code evidence — catalog-driven, provenance-tracked, and indepe
Lane Verify
Out-of-domain verification lane. Independently REFUTE other lanes' done-claims, defaulting to refuted when unc
Forecost
Local, content-free flight recorder for AI agents: independently meter usage, reconcile vendor costs, and enfo