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.
npx agentmods add instructions/actava-ai/chi-bench/claude-mdgit clone --depth 1 https://github.com/actava-ai/chi-benchWrote 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/instructions/actava-ai/chi-bench/claude-md)<a href="https://agentmods.dev/instructions/actava-ai/chi-bench/claude-md"><img src="https://agentmods.dev/badge/instructions/actava-ai/chi-bench/claude-md.svg" alt="Measured on agentmods" 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 | $0.04421 | $0.04421 |
| Opus 5 | $0.02210 | $0.02210 |
| Sonnet 5 | $0.00884 | $0.00884 |
| Haiku 4.5 | $0.00442 | $0.00442 |
Grade A, and why
chi-bench CLAUDE.md scanned grade A with 1 finding 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 5d 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.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
Anthropic Console setting, not a harness/config bug. Reproduce with a bare curl before How it starts
The opening of the file, as written. The whole thing — 273 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
Self-improvement protocol
This file is a living document. Whenever a session in this repo would have gone smoother with a note here, edit CLAUDE.md before ending the turn — don't wait to be asked. Future sessions inherit this file but inherit nothing else from this one.
Triggers — add or update an entry when you notice any of these:
- The user corrects an approach ("no", "not that", "stop doing X", "again, like I said before"). → capture the rule and why (the reason the user gave, often a past incident).
- The user says "again" or you re-discover something already explained earlier in the session or in a prior session. Rediscovery = missing note.
- You hit the same error, friction point, or wrong-first-guess that has come up before (within this session or across sessions). One repeat is enough; don't wait for three.
- A command, path, env var, or convention was non-obvious and required grep / multiple file reads to locate.
- The user validates a non-obvious choice ("yes exactly", "that was the right call", accepting an unusual approach without pushback). Capture confirmations too — without them, future-you drifts away from validated approaches and re-litigates them.
- A command in the Commands section turned out to be wrong, stale, or missing a flag.
How to update:
- Add the note to the most relevant existing section (Commands, Things to remember, Architecture, Subdirectory note) rather than spawning new sections.
- Format: one-line rule first, then Why: clause when the reasoning is non-obvious, then optionally How to apply: when the trigger condition isn't obvious from the rule.
- If an existing item is wrong or stale, fix or delete it — don't accumulate cruft. CLAUDE.md is read in full at the start of every session; entries past line ~200 lose weight.
- Keep entries actionable and specific. "Be careful with X" is useless; "Run
foobeforebar, otherwise SQLite is left in state Y" is useful. - Mention CLAUDE.md updates in your end-of-turn summary so the user can review the diff.
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.
- 5d ago First seen · 273 lines · 4,421 tokens per session scan A a51181512fa0
chi-bench CLAUDE.md is an instructions file published in the GitHub repository actava-ai/chi-bench (59 stars, last pushed 3d ago), licensed Apache-2.0. It adds 4,421 tokens to every session, about $0.0221 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
benchmark AGENTS.md
AGENTS.md instructions for google/benchmark, a project described as: A microbenchmark support library.
mobilegym AGENTS.md
Instructions for Purewhiter/mobilegym, covering agents.md, project overview, type-checking strategy, eslint and with data graph generation.
ClawBench AGENTS.md
AGENTS.md instructions for TIGER-AI-Lab/ClawBench, covering clawbench -- agent context, what this is, project structure, setup and 2. configure at least one model.
strix-halo-guide AGENTS.md
AGENTS.md instructions for hogeheer499-commits/strix-halo-guide, covering agents.md, core project lens, documentation style, do-not-invent rules and benchmark claim rules.
openbench AGENTS.md
Instructions for minghinmatthewlam/openbench, covering openbench — agent context, local execution context, what openbench is, execution ownership and product goals (the two things we are building toward).
aws-bench AGENTS.md
AGENTS.md instructions for aws-bench/aws-bench, covering agents.md, setup, commands, project structure and code style.