Harness Novelty Checker — Development agent for Claude Code
Fresh-context duplicate detector for the benchmarking climb loop.
How to install Harness Novelty Checker
Installs to ~/.claude/agents/leadgrowgtm-loop-engineer-harness-novelty-checker.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/LeadGrowGTM/loop-engineer/HEAD/.claude/agents/harness-novelty-checker.md -o ~/.claude/agents/leadgrowgtm-loop-engineer-harness-novelty-checker.md Restart Claude Code, or start a new session, for it to be picked up.
What Harness Novelty Checker does
name: harness-novelty-checker description: Fresh-context duplicate detector for the benchmarking climb loop. Diffs a proposed variant against the variant ledger and rejects near-duplicates of already-measured variants BEFORE a measurement is spent. Did NOT invent the variant. Returns a binary NOVEL | DUPLICATE verdict citing the closest ledger match. Use as the second pre-measurement check in every climb cycle, after in-bounds. tools: Read, Glob model: claude-sonnet-5
You are the Harnes
Alternatives in Development
- Eval Engineer — GAIA evaluation framework specialist 1.5k ★
- Duplicate Detector — Internal agent for finding and consolidating duplicate or near-duplicate memories, rules, and learnings 150 ★
- Peer Reviewer Demographics — A simulated peer reviewer specializing in population representativeness, demographic measurement, and populati 134 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Harness Inbounds Checker
Fresh-context invariant enforcer for the benchmarking climb loop. Diffs a proposed variant against the declare
Perf Engineer
Performance diagnosis specialist for the Hydraia pipeline. Measurement-first — establishes a baseline, profile
Code Patterns Strategist
Read-only reuse and consolidation strategist. Use BEFORE writing a helper, schema, service, or utility (finds
Implement Variant Generator
Generates 3-5 implementation variants using different coding paradigms for
Variant Extensibility Reviewer
Reviews a stack built around a registry of variants (job types, executors, plugins, discriminated unions) for
Evidence Reconciler
Reconciles OncoKB vs CiVIC evidence for a single somatic variant and emits a unified AMP/ASCO/CAP tier call wi
Related Skills
Benchmarking Loop
Run a benchmark spec through the benchmarking loop (sweep or climb) and write a ranked variant ledger
Propose Rules
Review pending memory rule and principle proposals from the digital-twin pushback detector. Approves or reject
Colors
Run a color-only design audit — contrast, near-duplicates, dark mode, semantic coverage. Scoped variant of /ui