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/karanb192/bouncer/agents-mdgit clone --depth 1 https://github.com/karanb192/bouncerWrote 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/karanb192/bouncer/agents-md)<a href="https://agentmods.dev/instructions/karanb192/bouncer/agents-md"><img src="https://agentmods.dev/badge/instructions/karanb192/bouncer/agents-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 | $0.00461 | $0.00461 |
| Opus 5 | $0.00230 | $0.00230 |
| Sonnet 5 | $0.00092 | $0.00092 |
| Haiku 4.5 | $0.00046 | $0.00046 |
Grade F, and why
bouncer AGENTS.md scanned grade F with 5 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 3d 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.
Asks for rootmediumPrivilege escalation
A mod that escalates privileges can change anything on the machine, not only the project.
- `chmod 777`, `chown /`, `kill -9 1`, fork bombs, `npm publish`, `crontab -r` Reaches for credential fileshighPrivilege escalation
SSH keys, cloud credentials, git-credentials, .npmrc, /etc/shadow: reading these is how a config file becomes a credential leak.
- overwriting `/etc/passwd`, `/etc/shadow`, `/etc/sudoers`, `/etc/hosts` Downloads and executes remote codehighSupply chain
curl | sh runs whatever the server returns today, which is not necessarily what it returned when this was reviewed.
- `curl … | sh` (piping a URL straight to the shell); `curl`/`wget` to paste / exfil hosts Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- `rm -rf ~`, `rm -rf /`, `rm` on system dirs or with parent-traversal / glob targets Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
- `curl … | sh` (piping a URL straight to the shell); `curl`/`wget` to paste / exfil hosts What it actually says
Bouncer is active
A door-guard sits in front of your shell. Every run_shell_command call passes
the door before it runs; destructive commands are blocked by an enforced
BeforeTool hook, not by you, and not negotiable.
Don't try to run the footguns below. They will be bounced with the exact rule that fired, and the turn is wasted. Reach for the safe alternative instead.
Bounced at the door (non-exhaustive):
rm -rf ~,rm -rf /,rmon system dirs or with parent-traversal / glob targetsdd/mkfs/wipefsto a disk device; writing to/dev/sd*,/dev/nvme*- overwriting
/etc/passwd,/etc/shadow,/etc/sudoers,/etc/hosts DROP TABLE/DROP DATABASE/TRUNCATE;DELETE/UPDATEwith noWHEREredis-cli flushall/flushdb,dropdbcurl … | sh(piping a URL straight to the shell);curl/wgetto paste / exfil hosts- uploading or
cat-ing secrets:.env,id_rsa,*.pem,*.key, credentials git push --forceto main/master,git reset --hard,git clean -f,git checkout .chmod 777,chown /,kill -9 1, fork bombs,npm publish,crontab -r
Walk right in: read-only inspection, scoped edits, git status/diff/add/commit,
tests, builds, a WHERE'd UPDATE, normal npm/docker/psql work.
This advisory is belt-and-suspenders. The real enforcement is the hook, but
not provoking it keeps your turns productive. Tune with
BOUNCER_LEVEL=critical|high|strict; disable with BOUNCER_OFF=1.
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.
- 3d ago First seen · 28 lines · 461 tokens per session scan F 146eb0f7ef83
bouncer AGENTS.md is an instructions file published in the GitHub repository karanb192/bouncer (5 stars, last pushed 16d ago), licensed MIT. It adds 461 tokens to every session, about $0.0023 per session on Opus 5. A static security scan graded it F with 5 findings (asks for root, reaches for credential files, downloads and executes remote code). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
Guardrails AGENTS.md
AGENTS.md instructions for NVIDIA-NeMo/Guardrails, covering agents.md, agent skills, quick rules, repository map and setup.
probity CLAUDE.md
Instructions for nizos/probity, covering probity, discipline and layout.
vibeguard AGENTS.md
AGENTS.md instructions for majiayu000/vibeguard, covering agent instructions, scope, start here, core rules and delivery policy.
Guardrails CLAUDE.md
Claude Code instructions for NVIDIA-NeMo/Guardrails, a project described as: NeMo Guardrails is an open-source toolkit for easily adding programmable guardrails to LLM-based conversational systems.
cc-safe-setup CLAUDE.md
Instructions for yurukusa/cc-safe-setup, covering project rules, safety, code style and git.
klaudiush CLAUDE.md
Claude Code instructions for smykla-skalski/klaudiush, covering claude.md, project overview, commands, completion (shell completion scripts) and doctor (diagnose setup and configuration).