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/arnarg/crushout/agents-mdgit clone --depth 1 https://github.com/arnarg/crushoutWhat 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 | $0.01426 | $0.01426 |
| Opus 5 | $0.00713 | $0.00713 |
| Sonnet 5 | $0.00285 | $0.00285 |
| Haiku 4.5 | $0.00143 | $0.00143 |
Grade A, and why
crushout 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.
How it starts
The opening of the file, as written. The whole thing — 61 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md
What This Project Is
crushout is a Crush / Claude Code hook binary that auto-allows read-only bash commands. It reads a JSON payload from stdin (hook protocol), detects whether it's Crush or Claude Code format, parses the bash command with tree-sitter, checks every command against a rule set, and outputs a JSON allow/deny/prompt decision on stdout.
Commands
go test ./... # run all tests
go build ./cmd/crushout # build the binary
No Makefile, no CI config, no linter config, no other tooling.
Architecture & Data Flow
stdin (hook JSON, Crush or Claude Code format)
→ main.go: hook.Decode() detects protocol, skip non-bash/empty commands
→ config.Load(): load global + repo config, merge with defaults
→ bash.Parse(): tree-sitter parses bash into []Command
→ checker.Check(): iterate commands, track cd, apply rules
→ stdout: JSON Output (protocol-specific allow/deny/prompt)
Packages
| Package | Purpose |
|---|---|
cmd/crushout |
Entrypoint. Reads stdin JSON, loads config, wires checker, writes stdout JSON. |
internal/hook |
Hook interface, CrushInput/ClaudeInput structs, Decode() auto-detects protocol from field names. |
internal/bash |
Tree-sitter–based bash parser. Extracts Command structs from a bash AST. Detects "complex" constructs (command substitution $(), subshells (), process substitution <(), arithmetic $(())) and parse errors, both will not auto-allow. |
internal/checker |
Orchestrates parse + rule check. Tracks cd to ensure the working directory stays within RootDir. Resolves ~/... paths against HomeDir. Any cd that would escape the root causes a deny. |
internal/config |
Layered config loading. Reads global ($XDG_CONFIG_HOME/crushout/crushout.{yml,yaml}) and repo (<rootDir>/.crushout.{yml,yaml}) config, deep-merges both over built-in defaults. Repo config wins over global. Supports overwrite_defaults: true per-layer to replace the accumulated base entirely. |
internal/rules |
Recursive rule engine (Rule struct with Subcommands, PromptFlags, AllowFlags, Default). defaults.go contains the full built-in ruleset mapping command names to their allow/deny rules. |
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.
- 2d ago First seen · 61 lines · 1,426 tokens per session scan A fb178a14bc9b
crushout AGENTS.md is an instructions file published in the GitHub repository arnarg/crushout (2 stars, last pushed 1mo ago), licensed MIT. It adds 1,426 tokens to every session, about $0.0071 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.
Other instructions, from other repositories
agent-plugins-skills copilot-instructions.md
Instructions for richfrem/agent-plugins-skills, covering copilot instructions for agent-plugins-skills, 1. think before coding, 2. simplicity first, 3. surgical changes and 4. goal-driven execution.
ralph CLAUDE.md
Instructions for SantanderAI/ralph, covering ralph — instrucciones del proyecto and paridad bash ↔ powershell (importante).
claude-code-statusline CLAUDE.md
Instructions for glauberlima/claude-code-statusline, covering claude.md, project overview, development commands, testing and run all tests (unit + integration).
auto-uv-env AGENTS.md
Instructions for ashwch/auto-uv-env, covering agents.md - the auto-uv-env story, what this repo is, architecture diagrams, runtime activation flow and repository lifecycle.
kleosrules AGENTS.md
Instructions for kleosr/kleosrules, a project described as: Cursor harness pack: user rules, skills, Bash hooks, local HANDOFF memory. macOS, Linux, Windows (WSL).
ccmux AGENTS.md
Instructions for max-listov/ccmux: Read CLAUDE.md in this repository root before making any change. It is the single source of truth for working here — repository conventions, what must never be published, the order in which work is finished, committed and released, and the runtime rules (Bun, never Node/npm).