parallel-audit

Audits a codebase across multiple quality dimensions using parallel read-only subagents and produces a vetted fix plan.

64|11|Updated May 24, 2026
One-click install
npx skills add https://github.com/AlexanderMattTurner/agent-glovebox --skill parallel-audit-alexandermattturner
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: parallel-audit
Source: https://github.com/AlexanderMattTurner/agent-glovebox/tree/main/.claude/skills/parallel-audit
Command: npx skills add https://github.com/AlexanderMattTurner/agent-glovebox --skill parallel-audit-alexandermattturner

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Broad codebase audits are slow and shallow when done in a single pass: one reviewer misses whole classes of defects, duplicates findings, and delivers spot fixes instead of structural eliminations. This Skill fans out parallel read-only subagents across dimensions like security, robustness, test realness, and UX, then independently confirms findings and consolidates them into a partitioned fix plan. ## Core Features & Use Cases - Parallel fan-out audit: Launches one read-only subagent per dimension and file area with disjoint scopes, each required to cite exact file lines, severity, violated property, and a class-closing eliminator fix. - Independent confirmation and dedupe: Re-reads cited lines to refute or confirm high-severity findings, merges duplicates, and reconciles findings against open PR diffs. - Partitioned fix plan with self-critique: Groups surviving findings into as few PRs as the footprint bound allows, with commits partitioned by subsystem, plus an attached critique of the plan. - Use Case: Ask for a whole-repo security and robustness audit before a major release; receive dozens of confirmed findings grouped into a sequenced multi-PR fix plan rather than a pile of unverified lint. ## Quick Start Ask the agent to audit the whole codebase across security, robustness, testing, and UX dimensions using parallel subagents and deliver a confirmed fix plan.

Frequently Asked Questions about parallel-audit

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I audit an entire codebase for security and robustness issues?

Use a parallel fan-out audit that launches one read-only subagent per quality dimension and file area, each with a disjoint scope. Findings are then independently confirmed by re-reading the cited lines, deduplicated, and grouped into a partitioned fix plan.

How to use subagents to find bugs across a whole repository?

Assign each subagent a non-overlapping file list and one audit dimension, and launch them concurrently in a single message. Require every finding to include exact file and line citations, quoted evidence, the violated property, and a structural fix that closes the whole defect class.

When should I use a parallel audit instead of a code review?

Use a parallel audit for breadth across the whole codebase when the deliverable is a fix plan, not edits. Use peer-review or code-review for a pending diff, and a single-pass security review for the current branch only.

Does the parallel audit make code changes automatically?

No, the audit itself is read-only and outputs a report plus a fix plan. Code changes happen only in the optional execution step, where the main session implements every partition and opens PRs, never the subagents.

What happens when a subagent hits a rate limit during the audit?

Resume the agent with a follow-up message to its agentId so it reuses accumulated context instead of restarting. Only fully restart an agent that died before doing meaningful work, roughly three or fewer tool calls.

How are audit findings verified before reaching the fix plan?

Every high or critical finding is independently confirmed by opening the cited file at the cited lines and checking the claim holds. Findings that fail this read are dropped or downgraded, and very large audits can use a second adversarial fan-out to refute findings.