agent-belt AGENTS.md

A guide for Agent Belt, a command-line system that evaluates AI coding agents through multi-turn scenarios and scored reports.

In plain words
What is it for?
Use it to install, diagnose, run, score, and report on command-line coding agents.
Why use it?
It explains the public command, setup checks, project boundaries, and standard verification commands before changes are made.

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/jfrog/agent-belt/agents-md
Clone the repo
git clone --depth 1 https://github.com/jfrog/agent-belt

Made for: Codex, OpenCode.

Per session 2,594 This file is loaded in full into every session.
When invoked 2,594 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 $0.02594 $0.02594
Opus 5 $0.01297 $0.01297
Sonnet 5 $0.00519 $0.00519
Haiku 4.5 $0.00259 $0.00259

Measured 2d ago against content hash 18de22e8b291, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

agent-belt 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 · 186 lines

How it starts

The opening of the file, as written. The whole thing — 186 lines — stays where its author put it; the contents beside it link to each section on GitHub.

agent-belt - Guide for AI Coding Agents

This file is the single entrypoint. Read it top-to-bottom and you have everything you need to make a non-trivial change to this repo.

1. What this is

A universal evaluation harness for headless CLI agents. It runs multi-turn scenarios against any agent that has a CLI, scores the results with a combination of rule-based checks and LLM judges, and aggregates them into reports.

The only public surface is the belt console script. Everything under src/belt/{commands,runner,scorer,aggregator}/ is internal - callers that aren't the CLI must go through documented agent/scorer extension points.

2. Setup & first verification

uv sync                        # creates .venv, installs locked dev deps
uv run belt doctor             # checks Python, agents, providers, env
make check                     # lint + test - same as CI

belt doctor is the fastest way to sanity-check the install: it verifies entry points resolve, agents are reachable, judge providers are configured, and the active clone is the one your belt command points at (a real foot-gun if you have multiple checkouts).

3. Where to start (by task)

Task Open this first
Add a new agent docs/glossary/PLUGGABILITY.md (Authoring an agent), then src/belt/agent/base.py
Modify an existing agent's behaviour src/belt/agent/<agent>.py (subclass BaseAgentAdapter)
Add a new scorer plugin docs/glossary/PLUGGABILITY.md (Authoring a scorer), then src/belt/scorer/base.py
Add a new rule-based check src/belt/scorer/rules/scorer.py (read neighbours first)
Change LLM judging or prompts src/belt/scorer/llm/scorer.py
Add an LLM provider src/belt/scorer/llm/backend.py (subclass BaseJudgeBackend)
Change scenario JSON shape src/belt/entities.py + docs/glossary/SCENARIOS.md (reference appendix)
Add a CLI flag src/belt/commands/<cmd>.py (argparse only)
Add a CLI subcommand New commands/<name>.py (thin) + new module under the right phase library
Change run-phase pipeline src/belt/runner/phases/
Change threshold/aggregation logic src/belt/aggregator/
Add a new exporter (CSV / JUnit / vendor plugin) docs/glossary/PLUGGABILITY.md (Authoring an exporter), then src/belt/exporter/base.py
Tune the export phase or chain on eval / aggregate src/belt/commands/export.py (registry + dispatch)
Author a scenario docs/glossary/SCENARIOS.md + examples/scenarios/

Read the full file on GitHub · 186 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 · 186 lines · 2,594 tokens per session scan A 18de22e8b291

Subscribe to this mod's changes

agent-belt AGENTS.md is an instructions file published in the GitHub repository jfrog/agent-belt (18 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 2,594 tokens to every session, about $0.0130 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.