NodeBenchAI CLAUDE.md

Project instructions for working on NodeBench AI with Claude Code, an AI coding tool. They describe coordination rules, important files, and the required process for testing and merging changes.

In plain words
What is it for?
Use them when developing, testing, reviewing, merging, or deploying the NodeBench AI project.
Why use it?
They reduce conflicts between people or agents editing the same files and help prevent untested or incorrectly deployed changes.

Instructions file

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 instructions/homenshum/nodebenchai/claude-md
Clone the repo
git clone --depth 1 https://github.com/HomenShum/NodeBenchAI
Per session 4,629 This file is loaded in full into every session.
When invoked 4,629 The same file — it is already loaded in full.
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 $0.04629 $0.04629
Opus 5 $0.02315 $0.02315
Sonnet 5 $0.00926 $0.00926
Haiku 4.5 $0.00463 $0.00463

Measured 2d ago against content hash d1320372d536, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

NodeBenchAI CLAUDE.md 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 2d 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.md · 201 lines

How it starts

The opening of the file, as written. The whole thing — 201 lines — stays where its author put it; the contents beside it link to each section on GitHub.

NodeBench AI — Claude Code Project Instructions

Agent coordination (read FIRST when other agents may be active)

AGENT_COORDINATION.md (repo root) is the live ledger of who is editing what right now and what backend contracts are ready to call. Before editing a hot file (public/proto/home-v5.html, backend/convex/events.ts, backend/convex/schema/eventsSchema.ts, the ScratchNode e2e specs): scan its Active claims, claim your region before you start, hand off new backend contracts there, and never convex deploy/deploy:prod out-of-band to shared prod (declare new shared-table fields v.optional and announce them instead). This prevents the cross-agent collisions — schema drift, file-churn — that caused a real prod incident during the ScratchNode session.

Deploy & merge workflow (P0 — read before merging anything)

The repo's branch protection now enforces:

  • Required CI checks on every PR: CI / Typecheck, CI / Runtime smoke, CI / Build
  • enforce_admins: true — admin merges DO NOT bypass these checks anymore

Use gh pr merge <N> --auto --squash --delete-branch, NOT --admin --squash.

  • --auto waits for required checks then auto-merges. Same speed as --admin for green PRs.
  • --admin will be rejected by the platform on any PR that hasn't passed CI.
  • If you legitimately need to merge a CI-failing PR (rare), the user must lower the gate manually.

Vercel deploy webhook: the GitHub→Vercel push integration silently dropped events earlier (project link's sourceless: true flag). It was re-wired via vercel git disconnect && vercel git connect. As belt-and-suspenders, .github/workflows/vercel-deploy-hook-backup.yml curls a Deploy Hook URL on every push to main. Required repo secret: VERCEL_DEPLOY_HOOK_URL (already set). If this drift recurs, run gh secret list to confirm the secret is still there + check the workflow's last run for non-2xx.

If a deploy is stuck: prefer vercel build --prod && vercel deploy --prebuilt --prod over vercel --prod (latter uploads CWD's Windows-locked package-lock and breaks on sharp linux-x64). vercel redeploy <url> rebuilds the SAME commit, not current main — useless for getting recent merges live.

Read the full file on GitHub · 201 lines

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. 2d ago First seen · 201 lines · 4,629 tokens per session scan A d1320372d536

Subscribe to this mod's changes

NodeBenchAI CLAUDE.md is an instructions file published in the GitHub repository HomenShum/NodeBenchAI (14 stars, last pushed 19d ago), licensed MIT. It adds 4,629 tokens to every session, about $0.0231 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.