skills AGENTS.md

A set of project instructions for working with the arcjet/skills repository, including agent instructions, test runs, result aggregation, and transcript collection. It records practical details that are easy to miss when evaluating skills.

In plain words
What is it for?
Use it when running skill evaluations, preparing prompts for test agents, collecting agent transcripts, or checking aggregated grading results. It is especially relevant when organizing repeated runs and comparing configurations.
Why use it?
It prevents test results from being lost or misread because files are stored in the wrong place or configurations are ordered unexpectedly. It also helps preserve the records needed to inspect what subagents did.

Instructions file for CodexOpenCode

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/arcjet/skills/agents-md
Clone the repo
git clone --depth 1 https://github.com/arcjet/skills

Made for: Codex, OpenCode.

Per session 688 This file is loaded in full into every session.
When invoked 688 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 $0.00688 $0.00688
Opus 5 $0.00344 $0.00344
Sonnet 5 $0.00138 $0.00138
Haiku 4.5 $0.00069 $0.00069

Measured 2d ago against content hash 6504c220d8dd, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

skills AGENTS.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.

AGENTS.md · 33 lines

How it starts

The opening of the file, as written. The whole thing — 33 lines — stays where its author put it; the contents beside it link to each section on GitHub.

AGENTS.md

Loose lessons learned while using the skill-creator skill (at .agents/skills/skill-creator/). Terse on purpose. Always read .agents/skills/skill-creator/SKILL.md first — this file only supplements it.

Subagent prompts for test runs

  • Keep prompts minimal. The prescribed format (Skill path / Task / Input files / Save outputs to) is enough — don't add framing like "you're running an eval," "report what you did," or environment hints the agent should discover itself (e.g. that the CLI is pre-authenticated).
  • Don't add restrictions like "don't run the server" or "typechecking optional." Those bias the agent away from the verification step you're trying to measure.

Aggregation gotchas

  • The aggregator expects per-run subdirectories: eval-X/<config>/run-N/grading.json (plus timing.json). If you save outputs at eval-X/<config>/grading.json (no run-N), it picks up nothing and reports 0%.
  • Config ordering in benchmark.md is alphabetical, which means new_skill / old_skill and with_skill / without_skill flip the delta sign. Sanity-check the sign by hand.

Capturing subagent transcripts

  • Each Agent task writes its JSONL transcript to /tmp/claude-*/-workspaces-*/<session>/tasks/<task_id>.output. Don't cat or Read it — it's the full transcript and will blow your context.
  • Do cp it into the run directory as transcript.jsonl once the task completes. The viewer also looks for transcript.md at the same level.

Eval viewer

  • It embeds every file under outputs/ into a single inline <script>. Without an exclude list it'll happily inline 482 MB of node_modules, and a stray backtick inside a vendored file crashes the page. The current code excludes node_modules, __pycache__, .venv, .next, dist, build, lockfiles, etc. — extend that list if a new ecosystem shows up.
  • Bind to 0.0.0.0 (not 127.0.0.1) so devcontainer / remote setups can reach it via the forwarded port.

When the same baseline serves multiple iterations

Read the full file on GitHub · 33 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 · 33 lines · 688 tokens per session scan A 6504c220d8dd

Subscribe to this mod's changes

skills AGENTS.md is an instructions file published in the GitHub repository arcjet/skills (2 stars, last pushed 2d ago), licensed Apache-2.0. It adds 688 tokens to every session, about $0.0034 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 2026-08-31.

Related

Other instructions, from other repositories