Borrowing it
Nothing to install: this file belongs to saadnvd1/agent-os. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/saadnvd1/agent-os/main/.claude/agents/review-security.mdgit clone --depth 1 https://github.com/saadnvd1/agent-osWrote 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/agents/saadnvd1/agent-os/review-security)<a href="https://agentmods.dev/agents/saadnvd1/agent-os/review-security"><img src="https://agentmods.dev/badge/agents/saadnvd1/agent-os/review-security/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/saadnvd1/agent-os/review-security"><img src="https://agentmods.dev/badge/agents/saadnvd1/agent-os/review-security.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00081 | $0.02326 |
| Opus 5.5 | $0.00032 | $0.00930 |
| Sonnet 5.5 | $0.00016 | $0.00465 |
| Haiku 4.5 | $0.00008 | $0.00233 |
Grade A, and why
review-security 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 16d 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.
Runs shell commandslowCapability
Expected in a hook, worth knowing in a rule or an instructions file.
- `exec`/`execSync`/`spawn(..., { shell: true })` with any interpolated value: branch names, paths, PR titles, session names, commit messages. Use `execFile`/`spawn` with an argument array. How it starts
The opening of the file, as written. The whole thing — 124 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Security review
AgentOS runs coding agents with the user's shell, keys and repositories behind one HTTP server. Anyone who gets past its gate, or who can make it crash, owns or stops every session. Flag high-confidence holes in code introduced or modified by this branch, or old code this branch makes reachable (a new route to an old function counts).
Read first: lib/security/auth.ts (proxied, isPublicPath, networkTrust, authorize), lib/security/gate.ts (gateRequest, gateUpgrade), lib/security/net.ts, lib/security/route-guard.ts, and how server.ts calls them for HTTP and for upgrades.
Process
- Changed files: use the list in your prompt. If none was given: with a PR number,
gh pr diff <NUMBER> --name-only; otherwisegit diff main...HEAD --name-status -Mplusgit status --short --untracked-files=all. Read new files whole andgit diff main -- <file>for changed ones. - For each changed route, upgrade path, CLI, subprocess call, file write or prompt, trace the input back to where it enters (request, header, cookie, WebSocket message, env, argv, a file another session wrote, a PR body, a card) and forward to where it's used.
- Check every rule below.
- Report.
Rules
The gate: every way in goes through it
- A new HTTP path or WebSocket upgrade that doesn't pass
gateRequest/gateUpgrade, or an addition toisPublicPaththat isn't an exact path (a prefix match lets/public/../api/xor//api/xthrough; the static prefixes reject..,//, backslashes and percent-encoding, and new ones must too). - Trust decided by address alone. Loopback trust needs a loopback peer and a loopback Host and no proxy header.
proxied()must see every forwarding header: X-Forwarded-, Forwarded, Via, CF-, X-Real-IP, X-Client-IP, True-Client-IP, X-Original-Forwarded-For, Tailscale-*. A new reverse path (Connect's tunnel, tailnet HTTPS, a dev proxy) that arrives on loopback is remote, never loopback. Connect's tunnel listener (lib/connect/serve.ts) must mark every socket (markTunnelSocket) before any request, pin Host to the machine's own name, and drop X-Forwarded-Host, Forwarded and X-Original-Host; a marked socket never gets loopback, tailnet orAGENTOS_AUTH=off. - Tailnet trust from the 100.64.0.0/10 range alone. Only Tailscale's own interface (
tailscale0/utun) and the IPstailscalereports count; any other VPN uses that range too. AGENTOS_AUTH=offor any env switch that relaxes the gate applying to tunnelled (Connect) requests. It never does.- X-Forwarded-For or X-Forwarded-Proto honoured from anything but a loopback proxy, or Proto accepted as anything but exactly
http/https.
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.
- 16d ago Changed · +3 lines 55c5f0c4c480
- 22d ago First seen · 121 lines · 81 tokens per session scan A fcf0b638992e
review-security is an agent published in the GitHub repository saadnvd1/agent-os (11 stars, last pushed today), licensed MIT. It adds 81 tokens to every session and 2,326 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 1 finding (runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-07.
Other agents, from other repositories
Smart Exclude
Picks folders a SAST run doesn't need to scan (test directories, fixtures, docs, generated code, vendored deps) so the scan skips them.
atomic-auditor
Final gate for a finished implementation. Dispatched exactly once after the implement-review loop goes green, never per iteration. Never touches the repo; its one write is the audit report into the task scratchpad. Audits the delivered work as a whole: cumulative spec compliance, cross-iteration coherence…
Validate
Runs build, typecheck, lint and test; reports pass or fail with failure details; never fixes.
code-reviewer
You are a specialized code review subagent for Visor. Your role is to perform deep, comprehensive code reviews with a focus on.
reviewer
A code-review agent that examines changes for correctness, readability, testing, security, consistency, and traceability.
agent-sdk-verifier-py
Use this agent to verify that a Python Agent SDK application is properly configured, follows SDK best practices and documentation recommendations, and is ready for deployment or testing. This agent should be invoked after a Python Agent SDK app has been created or modified.