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/cai-layer/cai/copilot-instructionsgit clone --depth 1 https://github.com/cai-layer/caiWhat 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 | $0.02143 | $0.02143 |
| Opus 5 | $0.01071 | $0.01071 |
| Sonnet 5 | $0.00429 | $0.00429 |
| Haiku 4.5 | $0.00214 | $0.00214 |
Grade A, and why
cai copilot-instructions.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.
Same rule for `NSAppleScript` / `osascript`. Use the existing AppleScript escaper (backslash, quotes, newlines). A clipboard containing `"; do shell script "curl …"` is the canonical attack. How it starts
The opening of the file, as written. The whole thing — 143 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Cai — Copilot Review Instructions
Native macOS menu bar clipboard manager (SwiftUI + AppKit, macOS 14+, Apple Silicon). Privacy-first: no cloud, no telemetry.
Every rule here should be verifiable from a diff. If a rule fires, say why and cite the file/line. Be terse — one-line review comments, no essays.
Basics (merge-blocking)
- PR must build:
xcodebuild -scheme Cai -configuration Debug build - PR must pass tests:
xcodebuild -scheme Cai -configuration Debug test - 40+ content detection tests in
CaiTests/ContentDetectorTests.swift— these are the regression net
🔴 Security & privacy (merge-blocking)
Never log clipboard text, LLM prompts/output, API keys, or paths under ~
Flag any new print(…), NSLog(…), os_log(…), CrashReportingService breadcrumb, or SentrySDK.capture* call that includes these values — even in debug branches. The product ships with "no telemetry" as a core promise; one print("clipboard: \(text)") reaching Console.app breaks it.
Never interpolate untrusted text into a shell invocation
Flag "/bin/zsh -c \"\(text)\"" and any concatenation of clipboard/LLM text into shell command strings. Shell templates must go through the existing single-quote escaper in OutputDestinationService / ActionListWindow.runShellCommand. Raw text as a literal element in Process.arguments is fine; raw text inside -c is not.
Never interpolate untrusted text into AppleScript source
Same rule for NSAppleScript / osascript. Use the existing AppleScript escaper (backslash, quotes, newlines). A clipboard containing "; do shell script "curl …" is the canonical attack.
Deeplink substitution must use .urlQueryAllowed
Not .urlPathAllowed or .urlFragmentAllowed. Flag addingPercentEncoding(withAllowedCharacters:) with a wider set.
Webhook/deeplink URLs must stay HTTPS
Enforced in ExtensionParser and the extensions repo CI. Don't weaken or add an allowInsecure escape hatch without a separate security review. HTTPS doesn't block SSRF to localhost.* cert-holders — if a PR adds new webhook destinations, consider whether private-IP blocking is warranted.
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 · 143 lines · 2,143 tokens per session scan A 166bb9fc6d67
cai copilot-instructions.md is an instructions file published in the GitHub repository cai-layer/cai (56 stars, last pushed 8d ago), licensed MIT. It adds 2,143 tokens to every session, about $0.0107 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.
Other instructions, from other repositories
openquack AGENTS.md
Instructions for larryxiao/openquack, covering agents.md, posture, before you start, workflow and pr template (the required shape).
PokeTokenBar CLAUDE.md
Instructions for chattymin/PokeTokenBar, covering poketokenbar — claude 프로젝트 지침, 참조 문서 (필요할 때 읽는다), 기여 언어 규약 (오픈소스 대비 — english first), 릴리스 (자연어 트리거) and 확장 규약 (새 프로바이더/툴 추가 시).
TokenBar AGENTS.md
Instructions for Nanako0129/TokenBar, covering tokenbar agent routing, read order, invariants, authorization boundary and handoff.
TokenBar CLAUDE.md
Instructions for Nanako0129/TokenBar: Read AGENTS.md first, then docs/knowledge/README.md and the task-specific canonical document it routes to. If .agent-local/CLAUDE.md exists, read it after the canonical documents as additive machine-local guidance only; it must not override the canonical architecture…
localvoxtral AGENTS.md
Instructions for T0mSIlver/localvoxtral, covering localvoxtral — agent guide, build & test — read this first on a non-mac dev box, proof culture — non-negotiable, test tiers — the short version and ci / shipping.
barkeep AGENTS.md
Instructions for iannuttall/barkeep, covering agent notes, product rules, repo map, commands and app structure.