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/tmustier/codex-computer-use-mcp/agents-mdgit clone --depth 1 https://github.com/tmustier/codex-computer-use-mcpWhat 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 | $0.00332 | $0.00332 |
| Opus 5 | $0.00166 | $0.00166 |
| Sonnet 5 | $0.00066 | $0.00066 |
| Haiku 4.5 | $0.00033 | $0.00033 |
Grade A, and why
codex-computer-use-mcp 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.
What it actually says
Repository guidance
This package is a thin transport adapter to OpenAI's official signed macOS Computer Use tools. The calling agent selects each official method and its arguments. Codex and macOS remain authoritative for tool behaviour, app access and platform permissions.
Scope
- Preserve all ten official methods and no-permissions behaviour. MCP exposes them as typed tools; Pi may expose the same methods through one composable
computer_use({ code })tool. - Do not add wrapper permission prompts, app or intent allowlists, action gates, risk classifiers, content inspection, alternate modes or model-driven planning.
- Do not reintroduce app rewriting, same-app locking or focus telemetry.
- Add adapter logic only for transport compatibility, zero-model-turn execution, retained official sessions, process lifecycle, packaging or a concrete user-visible bug.
- Keep signature and Team ID verification, isolated temporary state, model-turn rejection and process cleanup.
- Run model-authored Pi code outside Pi's main thread with real cancellation, and preserve emitted observations plus completed-call history when a batch stops partway through.
- Preserve compatible additional tool arguments and pass app selectors through unchanged.
Review
Review the exact production path and report material, reachable regressions. Do not block focused work on speculative hardening or malformed input that the official producer cannot emit.
A request to review does not authorize posting to GitHub. Show the user the exact proposed review, comment or issue text and get approval before posting it.
Verification
npm ci
npm run check
npm run check:pi
npm test
npm run build
Use benign applications for live acceptance.
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 · 32 lines · 332 tokens per session scan A cdcf869f6c9f
codex-computer-use-mcp AGENTS.md is an instructions file published in the GitHub repository tmustier/codex-computer-use-mcp (13 stars, last pushed 10d ago), licensed MIT. It adds 332 tokens to every session, about $0.0017 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 2026-08-30.
Other instructions, from other repositories
MacOS-MCP CLAUDE.md
Claude Code instructions for CursorTouch/MacOS-MCP, covering macos-mcp: project documentation for claude, project overview, tech stack, architecture and entry point (src/macosmcp/main.py).
hunch-mcp AGENTS.md
Instructions for PrithviSeran/hunch-mcp, covering agents.md — working on hunch, setup & tests, architecture — one engine, three faces, adding or changing a tool and the focus-free contract (do not regress).
hunch-mcp CLAUDE.md
Instructions for PrithviSeran/hunch-mcp: See AGENTS.md for how this codebase works and the gotchas that matter (the focus-free contract, AX-layer traps, the NSWorkspace frontmost trap, adding tools, the vision path, plus packaging/release). It's the canonical guide for coding agents; this file just points to it so…
dsh-record-replay AGENTS.md
Instructions for humblebanana/dsh-record-replay, covering agents, build and test, hard rules and layout.
OpenAra AGENTS.md
Instructions for Aradotso/OpenAra, covering openara agent guide, read first, working rules and where things live.
silenthand-cua AGENTS.md
AGENTS.md instructions for edosulai/silenthand-cua, covering graphify and public repo hygiene.