Lists multiple specific concrete actions: 'find duplicated code, dead imports, security issues, and complexity hotspots.' Also mentions parallel subagents for fast analysis, adding implementation detail.

Clearly answers both 'what' (technical debt detection and remediation—duplicated code, dead imports, security issues, complexity hotspots) and 'when' (session end, before merge, code review prep, with explicit trigger phrases listed).

Excellent coverage of natural trigger terms users would say: 'find tech debt', 'scan for issues', 'check code quality', 'wrap up session', 'ready to commit', 'before merge', 'code review prep'. These are realistic phrases users would naturally use.

While the specific triggers like 'find tech debt' and 'complexity hotspots' are distinctive, terms like 'check code quality', 'code review prep', and 'security issues' could overlap with linting skills, security scanning skills, or general code review skills. The scope is broad enough to potentially conflict with more focused tools.

This skill is extremely verbose at ~350+ lines. It explains concepts Claude already knows (what cyclomatic complexity is, what dead code is, what SQL injection is), includes extensive tables of well-known patterns, and provides lengthy report templates, CI/CD configs, and troubleshooting sections that bloat the context window significantly. Much of this could be in referenced files or omitted entirely.

While the skill provides concrete CLI commands and report templates, the actual implementation is aspirational rather than executable. Commands like `/techdebt --deep` and tools like `ast-grep` are referenced but there's no actual code that orchestrates the subagent spawning, no real scripts to run, and the subagent instruction template is a placeholder. The pre-commit hook references a non-existent `claude skill techdebt` command.

The 5-step workflow is clearly sequenced and well-organized, but validation checkpoints are weak. There's no explicit verification that subagents completed successfully, no error handling for when tools are missing, and the auto-fix mode lacks a validation step after applying changes to confirm nothing broke. The safety rules for auto-fix are good but insufficient as a feedback loop.

The skill references external files (references/patterns.md, references/severity-guide.md) which is good, but the main file itself is a monolithic wall containing detection patterns, language support tables, integration patterns, advanced usage, troubleshooting, and best practices that should be split into separate reference files. The Quick Start section is well-placed but the body contains far too much inline detail.