dep-inheritance

Dependency inheritance audit: for every direct dependency answers the questions nobody asks - why is it here (from actual usage sites), how deep is the coupling, how replaceable is it, and what is the

SKILL.md

GitHub

/deps-audit

For each direct dependency answers the inheritance questions nobody asks: What is it REALLY used for, how deep is the coupling, how replaceable is it, and what is the concrete exit plan.

What this is for

  • Dependency decisions ("can we remove/replace X?") otherwise take days of research. This skill delivers a living inheritance register.
  • Read-only skill. No CVE scan (use npm audit & co - not duplicated). No transitive dependencies in deep analysis, only counted.

PROTECTION RULE - never ~/.claude/

Read-only skill. Guard required if write mode added later.

What You Must Do When Invoked

During analysis, assign a confidence level to each finding: proven (confirmed by evidence), likely (strong signal, needs review), or suspected (weak signal).

If /deps-audit -help or /deps-audit -h (without further arguments) is invoked: output the ## Usage section unchanged and stop.

Otherwise follow these steps in order, skipping none.

Step 1 - Help check

If invoked with -help or -h, output the ## Usage section unchanged and stop.

Step 2 - Clarify target

Clarify -ProjectDir and optionally a focus on individual dependency names. Get confirmation.

Step 3 - Inventory

& "<SKILL_DIR>/scripts/deps-inventory.ps1" -ProjectDir "<path>" [-Only <names>]

No manifest found: cleanly report, stop.

Step 4 - Registry metadata (optional)

Network available and user not opposed:

& "<SKILL_DIR>/scripts/registry-meta.ps1" -Names <dependency-names> -Ecosystem npm

metaError for individual packages: continue without maintenance signal for the affected dependency, note in report - no abort. Note: the field lastRelease comes from the time.modified field of the registry, which does not necessarily mean exact "last publish date" (registry metadata can be updated without a new release) - label as approximation in the report.

Step 5 - Analysis

Per direct dependency:

  1. Purpose (from usage sites): what is it REALLY used for - the honest answer is often "for a single function". State confidence.
  2. Coupling depth: shallow (few locations, simple calls) / medium / deep (API types in own signatures, inheritance, config magic). Evidence: usageCount + characteristic locations.
  3. Risk: combination of coupling depth, maintenance signal (if registry metadata available), license anomaly (only hint, no legal advice), unusedDeclared.
  4. Before confirming any unusedDeclared: check config files (json/yaml/rc) - e.g. eslint plugins are often referenced only there, not in code. Only then classify as "truly unused".
  5. Replaceability + exit plan: name concrete alternative(s) (check stdlib replacement first), effort estimate (hours/days/weeks), the first 3 concrete steps of the exit.
  6. If > 50 dependencies: register table for all, detail analysis only top 15 by risk, rest on request.
  7. Evidence requirement: purpose/coupling always with locations; maintenance statements only with registry data (otherwise "no metadata available" - never guess).

Step 6 - Write report

File deps-inheritance-report.md in the current working directory:

  1. Summary - inventory, top 3 risks, quick wins (unusedDeclared).
  2. Register table - name, purpose, coupling, risk, replaceability.
  3. Detail sections per notable dependency.
  4. Open questions.

Step 7 - Summarize

State the report path, quick wins first.

Usage

/deps-audit                    # interactive, all direct dependencies
/deps-audit <dir>              # analyze project
/deps-audit <dir> <dep> [...]  # only specified dependencies
/deps-audit -help