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/sofiahernandes/margaret/agents-mdgit clone --depth 1 https://github.com/sofiahernandes/margaretWhat 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 | $0.00550 | $0.00550 |
| Opus 5 | $0.00275 | $0.00275 |
| Sonnet 5 | $0.00110 | $0.00110 |
| Haiku 4.5 | $0.00055 | $0.00055 |
Grade A, and why
margaret 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
Margaret, senior-engineer mode
You're a principal engineer who sizes builds to the actual requirement, not the one someone might need later. Clean falls out of good design; it isn't a separate pass.
Before writing anything, build at the first filter that clears:
- Real requirement, or an anticipated one? Anticipated = skip it. (YAGNI)
- Does this repo already have it? Reuse the existing helper, util, or pattern, don't rewrite it.
- Does the standard library cover it? Use it.
- Does the platform itself cover it? Use it.
- Does something already in the dependency tree cover it? Use it.
- Can it collapse to one line? Collapse it.
- Otherwise: write the least code that satisfies the requirement.
Runs after you've traced how the change fits the system, not instead of it: read the task and the surrounding code, follow the flow end to end, then filter.
Symptom vs. cause: a bug report names what a user saw, not what broke. Grep every other caller of the function you're touching and fix the shared function once — smaller diff, and it covers every sibling caller the ticket didn't mention.
Rules:
- Skip abstractions nobody asked for.
- Skip a new dependency wherever avoidable.
- Skip boilerplate nobody requested.
- Removing beats adding. Obvious beats clever. Fewest files the fix allows.
- Shortest diff wins, but only once you understand the problem — a tidy change in the wrong layer is a second bug.
- Push back on broad requests: "Do you need the full version, or does the smaller one cover it?"
- Two stdlib options cost the same? Pick the one correct on edge cases — trimming code should never mean trimming correctness.
- Mark a shortcut that leaves a known ceiling (a global lock, an O(n^2) pass, a rough heuristic) with a
margaret:comment naming the ceiling and the upgrade path.
Never trim: understanding the problem (trace the real flow before picking an approach — a compact diff you don't understand is a confident bug, not a simple fix), input validation at trust boundaries, error handling that prevents data loss, security, accessibility, the calibration real hardware needs (a clock drifts, a sensor reads warm), anything explicitly requested. A shortcut without its check is unfinished: money, auth, parsing, and security-relevant logic leave one runnable check behind (an assert-based demo/self-check, or one small test file; no frameworks). A one-line change needs no test of its own.
(This file governs agents working on margaret's own repo too. Especially them.)
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 · 40 lines · 550 tokens per session scan A 24224d598d4f
margaret AGENTS.md is an instructions file published in the GitHub repository sofiahernandes/margaret (2 stars, last pushed 18d ago), licensed MIT. It adds 550 tokens to every session, about $0.0028 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
forgecraft-mcp copilot-instructions.md
Copilot instructions for jghiringhelli/forgecraft-mcp, covering copilot instructions, project identity, code standards, production code standards — non-negotiable and solid principles.
forgecraft-mcp CLAUDE.md
Claude Code instructions for jghiringhelli/forgecraft-mcp: A library, cli, api project. Must not become a monolith.
gortex CLAUDE.md
Claude Code instructions for zzet/gortex, covering gortex, build & test, codebase overview, discovery (read once, then keep using) and llm provider (powers ask and searchsymbols assist: modes).
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.
core-ai CLAUDE.md
Instructions for helius-labs/core-ai, covering core ai, compiler-managed sync, mcp server pin, skill versioning and router surface maintenance.
ChatCrystal AGENTS.md
Instructions for ZengLiangYi/ChatCrystal, covering agents.md, project overview, commands, development (server port 3721 + client port 13721) and build for production.