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/mbanderas/maestro/agents-mdgit clone --depth 1 https://github.com/mbanderas/maestroWhat 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 | $0.02627 | $0.02627 |
| Opus 5 | $0.01314 | $0.01314 |
| Sonnet 5 | $0.00525 | $0.00525 |
| Haiku 4.5 | $0.00263 | $0.00263 |
Grade A, and why
maestro 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.
How it starts
The opening of the file, as written. The whole thing — 239 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md -- Maestro Orchestration Kernel
Optional discipline layer for AI coding agents. It is independent of
Maestro Frontier and is installed only with --with-discipline. The full
multi-agent protocol lives in
docs/orchestration.md, loaded on demand.
S0-S10 are stable identifiers.
Calibrated defaults, not rigid laws — apply judgment, scale to the task. Lead with the work; the S1 gate is a checkpoint just before your first edit, never your opening move. A rule that plainly does not fit yields to doing the task well — say so and proceed. Hard invariants are narrow: verification honesty (S7.3), surgical scope (S7.4), compression integrity (S8).
0. Quality Standard [ALWAYS]
Do the whole thing, right, with tests and docs scaled to the change. Search before building; test before shipping. Bar: genuinely done, within scope.
1. Decision Gate [ALWAYS]
Engage first — read the request, orient on the files it names. The gate is the checkpoint just before your first edit, when you know the real file count; not your opening move.
Output one verdict line —
Maestro · frontier <on|off> — files=<n> concerns=<m> -> single-agent — <reason> or
Maestro · frontier <on|off> — files=<n> concerns=<m> -> multi-agent — <trigger met>.
files = every file created or modified; concerns = distinct areas
(commands, core, config, docs, tests). Obvious single-agent work: a
one-line reflex. No edits before the verdict. The frontier badge =
engine state: frontier on (<mode>/<preset-or-model>) when armed, else
frontier off; on Claude Code the gate-reminder hook injects it.
Multi-agent triggers (ANY — check FIRST): 5+ files across 2+ concerns with structurally independent subtasks, >15 messages single-agent, adversarial review, multiple skill domains. File count alone is NOT a trigger — 5+ files in one coherent or sequential change stay single-agent. Multi-agent must be earned by real parallelism, context isolation, or adversarial review; absent one of those, single-agent. A met trigger still downgrades on >60% file overlap between subtasks, or <=3 files in one dependency chain.
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 · 239 lines · 2,627 tokens per session scan A 7966d81d2a31
maestro AGENTS.md is an instructions file published in the GitHub repository mbanderas/maestro (5 stars, last pushed 6d ago), licensed MIT. It adds 2,627 tokens to every session, about $0.0131 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 2026-08-31.
Other instructions, from other repositories
aghub AGENTS.md
Instructions for AkaraChen/aghub, covering aghub knowledge base, structure, where to look, conventions and rust.
cyrus copilot-instructions.md
Copilot instructions for cyrusagents/cyrus: Note, there is a need to maintain the use of '--print' when running the claude exec commands because that is what makes it non-interactive.
agentic-playwright skill-creator.instructions.md
Instructions for idavidov13/agentic-playwright, covering skill creator, communicating with the user, creating a skill, capture intent and interview and research.
agent-toolkit AGENTS.md
Instructions for ulises-jeremias/agent-toolkit, covering agents.md — ai agent contract, what this toolkit does, repository structure, operating rules and how to add a skill.
domain-experts CLAUDE.md
Instructions for wonsukchoi/domain-experts, covering domain experts — session bootstrap, non-negotiables when adding or editing roles, dual-graph context policy, mandatory: always follow this order and token usage.
agent-skills AGENTS.md
Instructions for helderberto/agent-skills, covering agents.md, repository overview, project structure, integration model and opencode integration.