
from claudefiles9
Run a 3-agent adversarial bug hunt (Hunter, Skeptic, Referee) to find and verify real bugs in a codebase.
Nit orchestrates a three-agent adversarial workflow to discover, challenge, and verify code issues. A Hunter agent gathers candidate findings, a Skeptic independently verifies or challenges each claim, and a Referee reconciles both outputs into a final verified report with triage-ready entries.
Use Nit when you need high-fidelity, automated bug discovery across a repository and want structured, reviewable findings rather than noisy heuristics. It is suited for security reviews, code health audits, and pre-merge checks where independent verification improves signal quality.
.claude/nitignore.md.Designed for multi-agent orchestration platforms and assistant frameworks that support subagents and filesystem tools (Claude subagents, Agent tool environments, or comparable assistant runtimes).
Nit is a multi-agent adversarial bug-finding skill that orchestrates three isolated sub-agents (Hunter, Skeptic, Referee) to identify and verify bugs in a codebase. No bundled scripts — the skill is purely prompt-driven, relying on Claude's Agent tool to launch sub-agents. The SKILL.md is well-structured with clear execution steps, triage support via .claude/nitignore.md, and a disciplined escalation flow. No security concerns found — it only reads code and produces reports, with no destructive actions or external network calls.
Clean skill — no scripts, no network calls, no destructive commands. Purely prompt-driven multi-agent orchestration. Well-structured with clear steps and triage workflow. Security is strong as the skill only reads code and generates reports. Architecture follows the skill spec well with frontmatter, references/, and prompts/ directories implied. Usefulness is moderate: bug hunting is valuable but the niche audience and Claude-specific tooling limit breadth.