Skip to content
Back to skills

Vcsdd Grading Criteria

ASecurity

Use this skill when evaluating VCSDD grading dimensions during adversarial review. Provides concrete evaluation checklists for each of the 5 dimensions and convergence signal detection.

  • 27 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added June 1, 2026
testingshell

Security analysis

A100/100

Scanned June 1, 2026

npx -y skills add sc30gsw/vcsdd-claude-code --skill vcsdd-grading-criteria --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Vcsdd Grading Criteria?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Vcsdd Grading Criteria
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/sc30gsw-vcsdd-grading-criteria/badge)](https://www.skillsdirectory.com/skills/sc30gsw-vcsdd-grading-criteria)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: vcsdd-grading-criteria
description: Use this skill when evaluating VCSDD grading dimensions during adversarial review. Provides concrete evaluation checklists for each of the 5 dimensions and convergence signal detection.
origin: VCSDD
---

# VCSDD Grading Criteria

## When to Activate
- During Phase 3 adversarial review
- Evaluating convergence signals in Phase 6
- Writing evidence blocks for verdicts

## Dimension Evaluation Checklists

### spec_fidelity
- [ ] Extract all REQ-XXX IDs from behavioral-spec.md
- [ ] For each REQ-XXX: find the corresponding test(s) in test files
- [ ] For each REQ-XXX: find the corresponding implementation
- [ ] Check for unspecified features in implementation

### edge_case_coverage
- [ ] Extract edge case catalog from behavioral-spec.md
- [ ] For each edge case: find the corresponding test
- [ ] Identify edge cases the spec MISSED (fuzzer thinking)
- [ ] Check boundary values (off-by-one, MAX_INT, empty collections)

### implementation_correctness
- [ ] Trace each function's logic against its spec requirement
- [ ] Check all error paths (what happens when dependency fails?)
- [ ] Look for null/undefined dereferences
- [ ] Check concurrent access assumptions
- [ ] Look for resource leaks (unclosed files, connections)

### structural_integrity
- [ ] Functions are focused (single responsibility)
- [ ] Names match their purpose precisely
- [ ] No copy-paste duplication
- [ ] Abstraction level consistent with surrounding code
- [ ] No dead code or commented-out blocks

### verification_readiness
- [ ] Pure functions take only primitives/data, return Result/Either
- [ ] Side effects confined to effectful shell modules
- [ ] Functions are small enough to reason about formally
- [ ] Proof obligations from verification-architecture.md are reflected in code structure

## Evidence Requirements

Every PASS verdict must include evidence blocks with:
- `type`: file | test-output | coverage | proof-result
- `location`: real file path (must be verifiable with ls)
- `description`: what was verified at that location

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.