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/anzellai/sky/claude-mdgit clone --depth 1 https://github.com/anzellai/skyWrote 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/instructions/anzellai/sky/claude-md)<a href="https://agentmods.dev/instructions/anzellai/sky/claude-md"><img src="https://agentmods.dev/badge/instructions/anzellai/sky/claude-md.svg" alt="Measured on agentmods" 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 | $0.09026 | $0.09026 |
| Opus 5 | $0.04513 | $0.04513 |
| Sonnet 5 | $0.01805 | $0.01805 |
| Haiku 4.5 | $0.00903 | $0.00903 |
Grade A, and why
sky CLAUDE.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 5d 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.
How it starts
The opening of the file, as written. The whole thing — 753 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
Entry point for Claude Code in the Sky repository. The agent-agnostic source of truth is AGENTS.md (imported below) — it teaches the language, the app-building decisions, the stdlib (via
sky doc), build & test, and the non-negotiable code rules for any AI tool. This file adds only the Claude-Code-specific operational rules that do not apply to other agents: the autonomous-loop protocol, mem-guard, and background-task hygiene.Read AGENTS.md first. Then the rules below govern how Claude Code operates when driving multi-step / autonomous work in this repo.
@AGENTS.md
Claude Code — operational rules
These are Claude-Code-specific. They sit on top of AGENTS.md's engineering norms
(no-deferral, regression-test-first, timeout-bounding, disk hygiene, template +
doc sync) — which apply to every agent — and add the loop/session mechanics that
are specific to Claude Code's tools (/loop, Judge/Architect agents, the
Workflow tool, ScheduleWakeup, Monitor, the file-based memory).
0. Goal fidelity in autonomous loops — INVIOLABLE
When the user gives an autonomous mandate (/loop AUTONOMOUS until <goal>, /loop AUTONOMOUS <goal>, or any equivalent multi-iteration
directive), the goal as the user worded it is the ONLY authority on
"done".
This rule applies to ALL autonomous mandates — current and future, v0.17 / v0.18 / every compiler-cycle close, every product mandate, every session resumed after compaction. It is structural: it survives compaction, new sessions, and any redefinition I might attempt under pressure.
The four hard rules
- The user's goal is captured VERBATIM at mandate start and
stored at
.claude/AUTONOMOUS_GOAL.mdin the project repo (so it survives clones, compactions, and new sessions). Subsequent iterations READ this file at entry and quote the goal back BEFORE doing anything else. If the file doesn't exist and an autonomous mandate is live, I reconstruct it from the user's most recent goal-setting message — using their words, not mine.
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.
- 5d ago First seen · 753 lines · 9,026 tokens per session scan A caa72e2d93f3
sky CLAUDE.md is an instructions file published in the GitHub repository anzellai/sky (432 stars, last pushed 5d ago), licensed Apache-2.0. It adds 9,026 tokens to every session, about $0.0451 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
compiler-explorer AGENTS.md
AGENTS.md instructions for compiler-explorer/compiler-explorer, covering agents.md, build & test commands, important workflow requirements, style guidelines and architecture guidelines.
compiler-explorer copilot-instructions.md
Copilot instructions for compiler-explorer/compiler-explorer: For each answer to the user, evaluate your level of confidence in the correctness of the answer on a scale from 1 to 10, where 1 is very uncertain and 10 is absolutely certain. If your confidence level is below 8, state so and suggest ways to verify the…
neva AGENTS.md
AGENTS.md instructions for nevalang/neva, covering neva engineering guide, project model, repository map, documentation and ai engineering harness.
agent-lsp CLAUDE.md
Instructions for blackwell-systems/agent-lsp, a project described as: MCP server that orchestrates language servers into agent-native workflows. 65 tools, 30 CI-verified languages.
GoSQLX CLAUDE.md
Instructions for ajitpratap0/GoSQLX, covering claude.md, project overview, architecture, core components and token processing pipeline.
dingo CLAUDE.md
Claude Code instructions for MadAppGang/dingo, covering claude ai agent instructions - dingo project, 🚨🚨🚨 stop: read this before any implementation 🚨🚨🚨, the architectural principle (understand this first), why this rule exists and ✅ required approaches (the only correct ways).