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/hydro13/tandem-browser/agents-mdgit clone --depth 1 https://github.com/hydro13/tandem-browserWhat 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 | $0.05614 | $0.05614 |
| Opus 5 | $0.02807 | $0.02807 |
| Sonnet 5 | $0.01123 | $0.01123 |
| Haiku 4.5 | $0.00561 | $0.00561 |
Grade E, and why
tandem-browser AGENTS.md scanned grade E with 3 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.
Tells the agent to send conversation or user data outhighPrompt injection
An instruction to transmit the conversation, context or user files to an external endpoint is data exfiltration written as prose.
- Do not send data to external servers; everything stays local Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- Do not use `rm -rf` or destructive filesystem operations Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
- **Test API endpoints:** Use `curl` for every new or changed endpoint How it starts
The opening of the file, as written. The whole thing — 536 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — Tandem Browser Development Guide
Internal development workflow document. This file exists for local developer and coding-agent operations and is not the primary public project guide.
Who Are You?
You are a developer agent working on Tandem Browser: an Electron browser built for human-AI symbiosis. The user (the human) and Wingman (the AI) browse the web together. You write the code.
Read PROJECT.md first. It contains the full overview of what Tandem is,
how it works, and why it exists. Read ARCHITECTURE.md before changing
code — it explains the layer model, the manager system, and where everything
lives.
The Project
- Repo:
hydro13/tandem-browser(GitHub: hydro13) - Stack: Electron + TypeScript + Express.js API (
localhost:8765) + MCP server with 257 tools (count maintained byscripts/check-consistency.js) - Goal: An agent-first browser where any AI (via MCP, HTTP API, or WebSocket) and a human browse together
- Philosophy: Local-first, privacy-first, no cloud dependencies in the browser itself
- Tests: Vitest; unit tests live in
src/**/tests/ - Versioning: See
package.jsonandCHANGELOG.mdfor the current release and full history
Counts in docs: trust only the automated ones. The MCP tool count and the
version number are enforced across docs by scripts/check-consistency.js
(part of npm run verify) — those you can trust. Any other count you find in
documentation (managers, files, modules) may have drifted; the authoritative
sources are src/registry.ts for the manager list, src/mcp/tools/ for the
MCP tool surface, and src/api/routes/ for the HTTP API surface. When a doc
and the code disagree, the code is right — fix the doc while you are there.
Project Structure
tandem-browser/
├── src/ # TypeScript application code (main process)
│ ├── main.ts # Electron main process entry
│ ├── registry.ts # ManagerRegistry — source of truth for managers
│ ├── bootstrap/ # Manager instantiation and wiring
│ ├── api/ # Express API server; routes in api/routes/
│ ├── mcp/ # MCP server; tools in mcp/tools/ (one file per domain)
│ ├── security/ # 8-layer shield (see ARCHITECTURE.md)
│ ├── extensions/ # Browser extension system
│ ├── snapshot/ # Accessibility tree with @refs
│ ├── platform/ # Platform adapters (macOS baseline, Windows track)
│ └── <domain>/ # ~50 single-responsibility domains, one dir each
├── shell/ # Browser UI (Electron renderer)
├── cli/ # tandem CLI (@hydro13/tandem-cli)
├── docs/ # Public site + project documentation
│ ├── implementations/ # Completed implementation plans
│ ├── plans/ # Not-yet-implemented design docs
│ ├── templates/ # design-template.md, manager-pattern.md
│ ├── research/ # Analyses and feature inventories
│ └── archive/ # Historical documents
├── scripts/ # Build, launch, and consistency scripts
├── skill/ # Agent skill file (SKILL.md)
├── tests/ # Smoke tests (unit tests live in src/**/tests/)
├── ARCHITECTURE.md # System structure — read before changing code
├── PROJECT.md # Product vision
├── AGENTS.md # This file
├── TODO.md # Current priorities
└── CHANGELOG.md
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 · 536 lines · 5,614 tokens per session scan E 52e8d2f62ed6
tandem-browser AGENTS.md is an instructions file published in the GitHub repository hydro13/tandem-browser (602 stars, last pushed 16d ago), licensed MIT. It adds 5,614 tokens to every session, about $0.0281 per session on Opus 5. A static security scan graded it E with 3 findings (tells the agent to send conversation or user data out, recursive force delete, 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
hronaut AGENTS.md
AGENTS.md instructions for hronaut/hronaut, covering hronaut contributor instructions, runtime boundaries, making changes, verification and releases and adjacent repositories.
vscode ai-customization.instructions.md
Architecture documentation for VS Code AI Customization view. Use when working in src/vs/workbench/contrib/chat/browser/aiCustomization.
vscode agentHostTesting.instructions.md
Architecture documentation for VS Code AI Customization view. Use when working in src/vs/workbench/contrib/chat/browser/aiCustomization.
chat-on-steroids AGENTS.md
Instructions for totec448-spec/chat-on-steroids, covering chat on steroids — the agent map, 1. the app in sixty seconds, 2. where the bugs actually are, name the identity, then find where it is lost and 3. what is authoritative.
rustwright AGENTS.md
Instructions for Skyvern-AI/rustwright, covering repository instructions, public boundary, pull-request review, local verification and browser launches.
stashbase AGENTS.md
Instructions for liliu-z/stashbase, covering agent maintenance contract for this repo, github access for this repository, temporary worktrees, electron launch environment and documentation route.