Claude Code Plugins

Community-maintained marketplace

Feedback

trace-guard

@tkersey/dotfiles
29
0

TRACE plus safety guardrails (unsoundness, invariants, footguns, complexity); use when code may crash, corrupt data, hide invalid states, invite misuse, or become hard to reason about.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name trace-guard
description TRACE plus safety guardrails (unsoundness, invariants, footguns, complexity); use when code may crash, corrupt data, hide invalid states, invite misuse, or become hard to reason about.

TRACE Guard

When to use

  • Crash risk, data corruption risk, or resource lifetime hazards.
  • "Should never happen" comments, nullable surprises, or validation sprawl.
  • Misuse-prone APIs, confusing parameters, or silent failure paths.
  • Tangled control flow, deep nesting, or cross-file hop fatigue.

Quick start

  1. Triage failure modes (crash > corruption > logic).
  2. Run Unsoundness scan and produce a concrete counterexample.
  3. Strengthen invariants (construction-time > runtime checks).
  4. Defuse footguns with safer signatures and tests.
  5. Reduce incidental complexity and report via TRACE.

Unsoundness scan

Goal: remove crash/corruption classes, not just symptoms.

Steps:

  • Trace nullables, lifetimes, concurrency, and resource ownership.
  • Give a concrete counterexample input or scenario.
  • Apply the smallest sound fix that removes the class.
  • State the new invariant.

Invariant strengthening

Goal: make illegal states unrepresentable.

Steps:

  • Name the at-risk invariant and current protection level.
  • Propose stronger construction-time or type-level guardrails.
  • Sketch before/after types or parsers.
  • Recommend a verification check (property test or assertion).

Footgun defusal

Goal: reduce misuse probability and impact.

Steps:

  • List hazards ordered by likelihood x severity.
  • Provide minimal misuse snippets with surprising behavior.
  • Offer safer signatures or naming (named params, typestate, richer types).
  • Add a test/assertion to lock the edge down.

Complexity mitigation

Goal: keep essential complexity, remove incidental complexity.

Steps:

  • Separate essential vs incidental complexity.
  • Propose flatten -> rename -> extract steps, ranked by impact/effort.
  • Provide a small code sketch when structural changes are needed.
  • Cite violated TRACE letters.

TRACE report

Deliverables:

  • Findings ordered by severity with file:line references.
  • Violated TRACE letters for each finding.
  • Minimal fix or mitigation and the new invariant (if any).
  • Open questions and required tests.

Activation cues

  • "crash"
  • "data corruption"
  • "should never happen"
  • "footgun"
  • "misuse"
  • "invariant"
  • "hard to reason about"
  • "too complex"