duyanta123/dsh-refactor-insight

English | 简体中文

A DSH skill plugin: turns codebase smells into an executable refactoring plan with locations, priorities, and dependency ordering — a structural health check, not a diff review.

Turn codebase smells into an executable, priority-ordered refactoring plan.

Positioning

Complementary by design: arch-doc looks at whole-repo structure (understanding the current state), existing code-review tools look at PRs/diffs (gating changes), and this plugin looks at whole-repo smells (outputting an action plan).

It answers:

• Which files, functions, and classes exceed healthy thresholds (length / nesting / method count)?
• Which modules are too coupled, and are there dependency cycles?
• How should findings be prioritized by severity and fix cost?
• In what dependency order should the refactoring proceed?

Boundaries (red lines): outputs a plan only — never modifies code automatically; high-risk refactoring items are flagged by the LLM and require human confirmation. The script only yields hard facts (deterministic scanning, all six rule thresholds overridable); severity / cost / suggested actions are added by the LLM after pruning false positives, marked as inference.

Capabilities (v0)

Six deterministic smell rules (zero-dependency heuristics):

• Oversized files (default > 400 lines)
• Long functions (Python indent blocks / JS-TS brace pairing / Go function declarations; body > 80 lines)
• Deep nesting (default max indent / brace depth > 5)
• God objects (> 10 class methods or class body > 300 lines / Go receiver method count)
• Highly coupled modules (reuses arch-profile --deps output; hub modules by in/out degree + Tarjan cycle detection)
• TODO/FIXME density (noise metric, reported per 1k lines)

Installation

As a DSH plugin (recommended):

dsh plugin --profile web add "github:duyanta123/dsh-refactor-insight#v0.1.3"

Or from npm:

npm install dsh-refactor-insight

Compatibility tiers: the standalone diagnostic scripts (refactor-smell.mjs / arch-profile.mjs) run on Node.js >= 18; as a DSH 0.1.5-rc.2 plugin it is verified with Node.js >= 22.19. Run npm run test:compat to execute an isolated-profile add, dump-config, and startup smoke test.

Local development: add "dsh-refactor-insight": "file:<local-path>/dsh-refactor-insight" to the profile's package.json and "dsh-refactor-insight" to the bundles array.

Quick Start

1. Use as a DSH skill

After installing, restart the profile and tell the agent:

Use refactor-insight to health-check /path/to/repo

The skill runs a four-phase runbook (intake → scan → diagnose → prioritize & orchestrate) and generates REFACTOR-PLAN.md following the docs/refactor-plan-template.md skeleton.

2. Run only the diagnostic CLI (zero dependencies, works without DSH)

# All six rules
node scripts/refactor-smell.mjs <repo_path>

# Reuse arch-profile's dependency output for coupling/cycle detection
node scripts/arch-profile.mjs <repo_path> --deps > deps.json
node scripts/refactor-smell.mjs <repo_path> --deps-json deps.json

CLI Options

Option	Default	Description
--max-lines <N>	400	Oversized file threshold (lines)
--max-func-lines <N>	80	Long function threshold (body lines)
--max-nesting <N>	5	Deep nesting threshold (indent/brace depth)
--max-methods <N>	10	God object threshold (class method count)
--max-class-lines <N>	300	God object threshold (class body lines)
--max-coupling <N>	4	High-coupling threshold (in/out degree)
--max-depth <N>	4	Directory scan depth (1–10)
--include-dirs <a,b>	-	Analyze only these directories
--exclude-dirs <a,b>	-	Extra excluded directories
--deps-json <file>	-	Reuse arch-profile --deps output for coupling/cycle detection

Sample output (excerpt):

{
  "language": "javascript/typescript",
  "smells": [
    {
      "rule": "god-object",
      "type": "god_object",
      "path": "src/lib/GodService.js",
      "name": "GodService",
      "methods": 12,
      "threshold": 10,
      "severity": 2,
      "cost": "M"
    }
  ]
}

Output

REFACTOR-PLAN.md: a structured refactoring plan with the fixed skeleton of summary → findings table (location/type/evidence/severity/cost/suggested action/risk) → priority ordering → dependency-ordered orchestration → appendix (rule thresholds, scan scope, generation time).

Safety Boundaries

• Plan only: no diffs generated, no automatic code changes; high-risk items require human confirmation.
• Read-only scanning: the diagnostic scripts have zero dependencies, no subprocesses, no network, and never modify the target repo's source.
• Facts vs. inference: the script emits hard facts; severity/cost/suggestions added by the LLM are always marked as inference.

Troubleshooting

No coupling/cycle findings? That rule needs a dependency graph as input: run node scripts/arch-profile.mjs <repo_path> --deps > deps.json first, then pass it via --deps-json deps.json (see Quick Start).

The findings include false positives? By design: the script emits deterministic hard facts, and false-positive pruning is the LLM's job in the runbook phase (checked one by one with added context); when using the CLI alone, judge by the severity and evidence fields.

Many TODO/FIXME density findings? That rule is a noise metric (per-1k-line counts only) and does not require action; it carries the lowest weight in plan ordering.

Old sessions won't open after upgrading the DSH host to 0.1.5.x? The Session format V3 migration is irreversible and is host behavior; back up session logs before upgrading the host (see the 0.1.3 entry in CHANGELOG.md).

Documentation

• docs/refactor-plan-template.md — the REFACTOR-PLAN.md output skeleton
• CHANGELOG.md — release notes
• PLUGIN-MAINTENANCE.md — repo maintenance runbook
• DSH-REFACTOR-INSIGHT-开发计划.md — design decisions and iteration history

License

MIT