Skip to main content
Search our AI Skills
< All Topics
Print

Vibe Coding Guardrails

Posted
Updated
ByPeter Westerman

name: vibe-coding-guardrails

description: Audit a project or feature build against the 15 vibe-coding pitfalls and produce a risk assessment with remediation actions. Use at the start of any new product, before a major feature build, or before a release to verify that minimum viable safeguards are in place.

Vibe Coding Guardrails

Instructions

Run a structured audit of a project against the 15 known vibe-coding pitfalls. Produce a risk scorecard and a prioritized remediation plan.

Audit checklist — verify each safeguard exists and is current:

# Safeguard Artifact to Check Pass Criteria
1 Requirements documented REQUIREMENTS.md or equivalent in repo Exists; has user stories with acceptance criteria; updated within last sprint
2 AI context file in place CLAUDE.md / .cursorrules at project root Exists; reflects current stack, conventions, constraints; updated this month
3 Dependency verification THIRD_PARTY.md or lock file committed Lock file committed; no unverified packages; last audit date recorded
4 Code ownership policy Code review gate active Every merge requires review; reviewer can explain the code they approved
5 Error-state coverage Error-handling checklist in Definition of Done DoD includes: empty state, network failure, invalid input, loading state, unauthorized access
6 Security baseline Security instructions in CLAUDE.md + no secrets in repo CLAUDE.md contains security block; .gitignore excludes credentials; no hardcoded keys
7 Architecture documented ARCHITECTURE.md or architecture section in docs Exists; matches actual file structure; updated on structural changes
8 Documentation freshness Docs updated on last feature change README reflects current behavior; API docs match endpoints; no stale references
9 Scope control Parking lot backlog exists In-scope items listed; out-of-scope explicitly declared; change request process defined
10 Code style enforced Linting/formatting config committed Config files present; pre-commit or CI enforcement active
11 Dependency audit Last audit date < 30 days npm audit / pip-audit / cargo audit run; results clean or exceptions documented
12 Deployment environment specified DEPLOY.md or deployment section in docs Target platform, runtime, env vars, and constraints documented
13 Staffing roles matched Roles assigned to build phases Requirements owner, architecture owner, QA lead, and reviewer identified
14 Tool redundancy Fallback workflow exists At least one alternative tool or manual workflow documented for core build tasks
15 Rollback procedure Rollback section in DEPLOY.md Trigger conditions, steps, verification, and communication plan documented

Risk scoring:

  • For each safeguard: Pass (0), Partial (1), Missing (2)
  • Total score: 0-6 = Green, 7-14 = Yellow, 15-30 = Red
  • Any single Critical pitfall (1, 2, 3, 15) scored Missing = auto-Red regardless of total

Output format:


## Vibe Coding Guardrail Audit — [Project Name]
**Date**: YYYY-MM-DD
**Overall Risk**: Green / Yellow / Red

### Scorecard
| # | Safeguard | Status | Score | Notes |
|---|-----------|--------|-------|-------|
| 1 | Requirements | Pass/Partial/Missing | 0/1/2 | [detail] |
...

### Total Score: [N] / 30 — [Green/Yellow/Red]

### Critical Gaps (fix before building)
[List any Missing items from pitfalls 1, 2, 3, 6, 15]

### High-Priority Gaps (fix this sprint)
[List any Missing items from pitfalls 4, 5, 7]

### Recommended Actions
1. [Specific action with file path and responsible agent/role]

Antigravity as QA evidence layer:

In the ITI three-lane toolchain (Cursor for development, Claude Code for context management, Antigravity for test/debug), Antigravity provides concrete evidence for several guardrail verifications:

  • Pitfall 5 (Error-state coverage): Use Antigravity’s browser sub-agent to capture screenshots of error states, empty states, and loading states. Browser recordings serve as auditable evidence that error handling was tested.
  • Pitfall 14 (Tool redundancy): The three-lane model itself satisfies this — Cursor and Claude Code provide fallback when Antigravity is unavailable, and vice versa.
  • Pitfall 15 (Rollback procedure): Antigravity’s Walkthrough artifacts document test session results. Include rollback verification test cases in Antigravity test dispatches.
  • Post-audit remediation: After the guardrail audit, dispatch Antigravity agents to verify that Critical and High-Priority gaps have been closed — the agent produces evidence as screenshots and test results.

See the antigravity-testing skill for agent dispatch patterns and the antigravity-browser-qa skill for browser-based evidence capture.

Outputs: Risk scorecard, prioritized remediation plan, safeguard status per pitfall.

Table of Contents
Privacy Policy Terms and Conditions Cookie Policy