tech-debt-audit

Tech Debt Audit

This skill runs a comprehensive technical debt audit by spawning 11 specialized agents in parallel, each examining the codebase from a different angle.

How It Works

1. Context gathering — reads CLAUDE.md, package.json, tsconfig.json, file tree, and git churn to understand the project
2. Parallel audit — launches 11 agents simultaneously, each focused on one category of technical debt
3. Aggregation — merges findings, deduplicates overlapping issues, boosts severity when multiple agents converge
4. Report — presents grouped issues ranked by severity
5. Issue creation — offers to create GitHub issues for selected findings

The 11 Audit Agents

Agent	Focus	Model
dependency-auditor	Outdated deps, CVEs, unused packages	haiku
code-quality-analyzer	Code smells, complexity, duplication, dead code	sonnet
architecture-reviewer	Circular deps, layer violations, pattern inconsistencies	sonnet
test-coverage-analyzer	Untested modules, test quality, integration gaps	sonnet
error-handling-auditor	Swallowed errors, missing error handling, resilience	sonnet
performance-analyzer	N+1 queries, missing indexes, unbounded queries	sonnet
security-scanner	Secrets, injection, validation, auth gaps	sonnet
documentation-auditor	Missing API docs, stale docs, missing ADRs	haiku
type-safety-auditor	any usage, unsafe assertions, strict mode gaps	sonnet
api-design-reviewer	Endpoint naming, pagination, error consistency	sonnet
complexity-reviewer	Over-engineering, unnecessary abstractions, YAGNI	sonnet

Severity Boosting

When multiple agents flag the same area or pattern, severity is bumped one level:

• An issue found by 1 agent keeps its original severity
• An issue found by 2+ agents gets bumped (LOW→MEDIUM, MEDIUM→HIGH, HIGH→CRITICAL)

Issue Grouping

Related findings are merged into single actionable issues:

• 8 endpoints missing pagination → one issue
• 15 files using any type → one issue
• 3 circular dependency chains → one issue

Usage

Run /tech-debt-audit to start a full codebase audit. The command handles everything automatically — context gathering, agent spawning, aggregation, and reporting.