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.
git clone --depth 1 https://github.com/athola/claude-night-marketWrote 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.
[](https://agentmods.dev/commands/athola/claude-night-market/karpathy-check)<a href="https://agentmods.dev/commands/athola/claude-night-market/karpathy-check"><img src="https://agentmods.dev/badge/commands/athola/claude-night-market/karpathy-check/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/commands/athola/claude-night-market/karpathy-check"><img src="https://agentmods.dev/badge/commands/athola/claude-night-market/karpathy-check.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00028 | $0.00578 |
| Opus 5 | $0.00014 | $0.00289 |
| Sonnet 5 | $0.00006 | $0.00116 |
| Haiku 4.5 | $0.00003 | $0.00058 |
Grade A, and why
karpathy-check 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 8d 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.
How it starts
The opening of the file, as written. The whole thing — 80 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Karpathy Check
A four-question pre-flight gate adapted from
observations on common LLM coding pitfalls. Loads the
imbue:karpathy-principles skill and walks each
principle as a checkable question.
Usage
# Run as an inline pre-flight before starting work
/imbue:karpathy-check
# Pass a task description to anchor the check
/imbue:karpathy-check Add rate limiting to the signup endpoint
What It Does
- Loads
Skill(imbue:karpathy-principles). - Walks the four-question battery:
- Think Before Coding: did I list assumptions, or guess silently?
- Simplicity First: would a senior engineer call this overcomplicated?
- Surgical Changes: does every changed line trace to the request?
- Goal-Driven Execution: can I prove this is done with a check, not a feeling?
- For each question, captures the answer and any blockers.
- Outputs a verdict: PASS, NEEDS CLARIFICATION, or REWORK.
Verdict Semantics
- PASS: all four answered concretely. Begin implementation.
- NEEDS CLARIFICATION: at least one assumption or ambiguity needs the user's input. Surface the question, do not guess.
- REWORK: the plan as drafted will violate one or more principles. Reshape before starting.
When to Use
- Before any non-trivial implementation
- When converting a vague request into a verifiable plan
- During code review, applied to your own diff
- When a task feels "too easy" (often a sign of hidden assumptions)
When NOT to Use
- Trivial typos and one-line fixes
- Throwaway scripts and exploratory spikes
- Documentation-only changes
- Production fires where the failing test wait is too long
See modules/tradeoff-acknowledgment.md in the skill
for the full boundary discussion.
Related
Skill(imbue:karpathy-principles)- the underlying four-principle synthesisSkill(imbue:scope-guard)- worthiness scoring before the four-question gateSkill(imbue:proof-of-work)- Iron Law TDD enforcement after the gate clears/imbue:justify- post-implementation audit using the same principles in reverse
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.
- 8d ago First seen · 80 lines · 28 tokens per session scan A acba201b12a2
karpathy-check is a command published in the GitHub repository athola/claude-night-market (337 stars, last pushed yesterday), licensed MIT. It adds 28 tokens to every session and 578 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 2026-09-03.
Other commands, from other repositories
analyze-codebase
Generate comprehensive analysis and documentation of entire codebase.
prompt
System instructions for writing effective prompts. Apply when generating commands, skills, agents, or any LLM instructions.
anti-slop-designer
Use when a UI looks machine-made rather than decided. Violet gradients, glassmorphism everywhere, identical cards in a grid, untouched shadcn or Material defaults, emoji bullets, or copy stuffed with seamless and unlock.
design-critic
Use when you want an honest read on a design instead of encouragement. What is wrong, ranked by severity, with the reason attached. Works on a screenshot, a mockup, or a described screen.
ui-copywriter
Use when writing marketing or landing copy. Headlines, hero copy, CTAs, value propositions and section copy that sound like a person making a specific claim rather than a template being filled in.
data-visualization-specialist
Use when a chart misleads or confuses. Wrong chart type for the data, a truncated axis, colour that encodes nothing, unreadable labels on mobile, or charts no screen reader can use.