VibeGuard: Skill Validate

VibeGuard: Skill Validate is a command for Claude Code from majiayu000/vibeguard. It costs 18 tokens per session (603 once invoked), scanned A, original, MIT.

A test gate for proposed coding-agent skills. It compares recorded tasks with and without a skill, including tasks the skill should not change.

In plain words
What is it for?
Use it to validate a new SKILL.md against baseline and held-out scenarios, then review success, failure, repair, and regression evidence.
Why use it?
It helps separate real improvements from guesses, repairs that cause regressions, outdated evidence, or effects on unrelated work.

Command for Claude Code

Written for Claude Code: argument-hint in frontmatter.

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add commands/majiayu000/vibeguard/skill-validate
Clone the repo
git clone --depth 1 https://github.com/majiayu000/vibeguard

Made for: Claude Code.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for VibeGuard: Skill Validate

README.md
[![agentmods](https://agentmods.dev/badge/commands/majiayu000/vibeguard/skill-validate.svg)](https://agentmods.dev/commands/majiayu000/vibeguard/skill-validate)
Your own site
<a href="https://agentmods.dev/commands/majiayu000/vibeguard/skill-validate"><img src="https://agentmods.dev/badge/commands/majiayu000/vibeguard/skill-validate.svg" alt="Measured on agentmods" height="20"></a>
Per session 18 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 603 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00018 $0.00603
Opus 5 $0.00009 $0.00302
Sonnet 5 $0.00004 $0.00121
Haiku 4.5 $0.00002 $0.00060

Measured 6d ago against content hash 4821c284a18b, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-05, from the pricing page.

Security

Grade A, and why

VibeGuard: Skill Validate scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude/commands/vibeguard/skill-validate.md · 57 lines

What it actually says

Core Concept

  • A proposed skill is an intervention, not just documentation.
  • Accept it only after measuring without_skill vs with_skill on recorded scenarios.
  • Keep facts separate from judgment: repairs, regressions, stale evidence, and unrelated-task regressions are explicit.

Minimum JSONL Schema

Each nonblank line in the baseline or held-out file is one scenario:

{
  "scenario_id": "incident-1",
  "scenario_type": "target | unrelated",
  "without_skill": {"outcome": "success | failure"},
  "with_skill": {"outcome": "success | failure"},
  "scored_against_agent": "claude-opus-4-7",
  "scored_at": "2026-05-18",
  "notes": "optional human judgment notes"
}

Steps

  1. Use /vibeguard:learn or a draft PR to propose a SKILL.md.
  2. Record 3-5 scenarios:
    • At least one motivating incident the skill should repair.
    • At least two unrelated tasks the skill should not affect.
    • Optional held-out scenarios for the final verdict.
  3. Run:
    python3 ~/vibeguard/scripts/skill_validate.py \
      --proposed-skill path/to/SKILL.md \
      --baseline-trajectories path/to/baseline.jsonl \
      --held-out path/to/held-out.jsonl \
      --current-agent <agent-or-model-id>
    
  4. Treat the verdict as the gate:
    • pass: repair > regression, at least one repair, no regressions.
    • needs_justification: repair beats regression but at least one regression needs a written trade-off.
    • advisory: unrelated task regression exists; do not install as a hard instruction.
    • stale: evidence was scored against a different or too-old agent/model.
    • fail: no demonstrated repair, or regressions are greater than/equal to repairs.
  5. Attach the emitted artifact path from ${VIBEGUARD_HOME:-${HOME}/.vibeguard}/artifacts/skill-validate/<project>/ to the PR or final answer.

Rules

  • Do not accept a new or changed skill on senior judgment alone.
  • Do not hide regressions in summary prose; report the count and affected scenario IDs.
  • Docs-only typo fixes can opt out; changes to constraints or decision rules need this gate.
Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 6d ago First seen · 57 lines · 18 tokens per session scan A 4821c284a18b

Subscribe to this mod's changes

VibeGuard: Skill Validate is a command published in the GitHub repository majiayu000/vibeguard (41 stars, last pushed yesterday), licensed MIT. It adds 18 tokens to every session and 603 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.