pi-agent-browser-native AGENTS.md

Instructions for a native pi integration of agent-browser, a command-line browser tool, without bundling the browser itself. They describe supported versions, documentation, commands, and how the integration should pass requests through.

In plain words
What is it for?
Using agent-browser from pi, checking its supported baseline, maintaining the wrapper, and documenting commands and tool-result categories.
Why use it?
They keep the integration aligned with the separate browser tool and avoid adding behavior that conflicts with it.

Instructions file for CodexOpenCode

Install

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.

agentmods
npx agentmods add instructions/fitchmultz/pi-agent-browser-native/agents-md
Clone the repo
git clone --depth 1 https://github.com/fitchmultz/pi-agent-browser-native

Made for: Codex, OpenCode.

Per session 15,663 This file is loaded in full into every session.
When invoked 15,663 The same file — it is already loaded in full.
Security scan A 1 finding. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.15663 $0.15663
Opus 5 $0.07831 $0.07831
Sonnet 5 $0.03133 $0.03133
Haiku 4.5 $0.01566 $0.01566

Measured 2d ago against content hash 57890b3bdeb5, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

pi-agent-browser-native AGENTS.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 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.

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

**Opt-in real upstream contract:** `npm run verify -- real-upstream` runs the sessionless plugin probe first, then the broad browser contract in a separate fail-fast `--test-force-exit` process, both from `test/agent-bro
AGENTS.md · 201 lines

How it starts

The opening of the file, as written. The whole thing — 201 lines — stays where its author put it; the contents beside it link to each section on GitHub.

pi-agent-browser-native

Native pi integration of agent-browser as a pi tool.

Product-specific guidance

  • Do not bundle agent-browser in this project.
  • Assume agent-browser is installed separately and available on PATH.
  • Baseline docs and verification against the current recommended agent-browser release. Runtime support uses the stable minimum in scripts/agent-browser-target.mjs (currently 0.35.0) with no upper cap and no version-specific compatibility shims.
  • Do not add backwards-compatibility shims for unsupported upstream versions.
  • Keep the integration thin and close to upstream agent-browser behavior.
  • Agent usability is the product boundary. Do not add wrapper-owned authorization or safety gates around upstream-supported commands, paths, sessions, state, restore, profiles, config, file access, launch arguments, or environment variables. Invoking agent_browser authorizes the requested upstream capability; pass it through and let upstream, the browser, and the operating system enforce their own contracts. Wrapper validation is for input shape, process integrity, lifecycle correctness, and truthful evidence—not for deciding what the agent or operator may do.
  • The primary UX is the agent invoking the native tool directly, not a slash-command-heavy manual workflow.
  • Do not overengineer or solve hypothetical problems that do not exist in observed behavior.
  • Do not add a reusable named "browser recipe" runtime layer on top of native job / qa / batch without new dogfood evidence and an explicit design pass; the current no-adopt decision and revisit criteria live in docs/ARCHITECTURE.md and docs/SUPPORT_MATRIX.md.
  • Thoroughly check official pi docs/examples/source behavior before inventing bespoke integration patterns. Prefer an official pi mechanism whenever one exists.
  • Prioritize the global install path first. Most users will install this extension globally, not as a project-local extension.
  • Keep the local extension-side documentation good enough that an agent can use the tool without relying on direct agent-browser binary help; when upstream agent-browser changes, update the repo-readable command reference, prompt guidance, README/docs, and any relevant tests in the same work.
  • For this repository, assume a single operator model: no human and no other agent is making changes here besides you.
  • Treat every lingering scratch file, temp artifact, browser session, tmux session, or other side effect related to this repository as your responsibility to clean up.

Read the full file on GitHub · 201 lines

Changes

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.

  1. 2d ago First seen · 201 lines · 15,663 tokens per session scan A 57890b3bdeb5

Subscribe to this mod's changes

pi-agent-browser-native AGENTS.md is an instructions file published in the GitHub repository fitchmultz/pi-agent-browser-native (205 stars, last pushed 3d ago), licensed MIT. It adds 15,663 tokens to every session, about $0.0783 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.